]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid10.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 2000-2004 Neil Brown | |
5 | * | |
6 | * RAID-10 support for md. | |
7 | * | |
25985edc | 8 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
9 | * |
10 | * | |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
5a0e3ad6 | 21 | #include <linux/slab.h> |
25570727 | 22 | #include <linux/delay.h> |
bff61975 | 23 | #include <linux/blkdev.h> |
056075c7 | 24 | #include <linux/module.h> |
bff61975 | 25 | #include <linux/seq_file.h> |
8bda470e | 26 | #include <linux/ratelimit.h> |
43b2e5d8 | 27 | #include "md.h" |
ef740c37 | 28 | #include "raid10.h" |
dab8b292 | 29 | #include "raid0.h" |
ef740c37 | 30 | #include "bitmap.h" |
1da177e4 LT |
31 | |
32 | /* | |
33 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
34 | * The layout of data is defined by | |
35 | * chunk_size | |
36 | * raid_disks | |
37 | * near_copies (stored in low byte of layout) | |
38 | * far_copies (stored in second byte of layout) | |
c93983bf | 39 | * far_offset (stored in bit 16 of layout ) |
1da177e4 LT |
40 | * |
41 | * The data to be stored is divided into chunks using chunksize. | |
42 | * Each device is divided into far_copies sections. | |
43 | * In each section, chunks are laid out in a style similar to raid0, but | |
44 | * near_copies copies of each chunk is stored (each on a different drive). | |
45 | * The starting device for each section is offset near_copies from the starting | |
46 | * device of the previous section. | |
c93983bf | 47 | * Thus they are (near_copies*far_copies) of each chunk, and each is on a different |
1da177e4 LT |
48 | * drive. |
49 | * near_copies and far_copies must be at least one, and their product is at most | |
50 | * raid_disks. | |
c93983bf N |
51 | * |
52 | * If far_offset is true, then the far_copies are handled a bit differently. | |
53 | * The copies are still in different stripes, but instead of be very far apart | |
54 | * on disk, there are adjacent stripes. | |
1da177e4 LT |
55 | */ |
56 | ||
57 | /* | |
58 | * Number of guaranteed r10bios in case of extreme VM load: | |
59 | */ | |
60 | #define NR_RAID10_BIOS 256 | |
61 | ||
34db0cd6 N |
62 | /* When there are this many requests queue to be written by |
63 | * the raid10 thread, we become 'congested' to provide back-pressure | |
64 | * for writeback. | |
65 | */ | |
66 | static int max_queued_requests = 1024; | |
67 | ||
e879a879 N |
68 | static void allow_barrier(struct r10conf *conf); |
69 | static void lower_barrier(struct r10conf *conf); | |
fae8cc5e | 70 | static int enough(struct r10conf *conf, int ignore); |
0a27ec96 | 71 | |
dd0fc66f | 72 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 73 | { |
e879a879 | 74 | struct r10conf *conf = data; |
9f2c9d12 | 75 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 76 | |
69335ef3 N |
77 | /* allocate a r10bio with room for raid_disks entries in the |
78 | * bios array */ | |
7eaceacc | 79 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
80 | } |
81 | ||
82 | static void r10bio_pool_free(void *r10_bio, void *data) | |
83 | { | |
84 | kfree(r10_bio); | |
85 | } | |
86 | ||
0310fa21 | 87 | /* Maximum size of each resync request */ |
1da177e4 | 88 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 89 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
90 | /* amount of memory to reserve for resync requests */ |
91 | #define RESYNC_WINDOW (1024*1024) | |
92 | /* maximum number of concurrent requests, memory permitting */ | |
93 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
94 | |
95 | /* | |
96 | * When performing a resync, we need to read and compare, so | |
97 | * we need as many pages are there are copies. | |
98 | * When performing a recovery, we need 2 bios, one for read, | |
99 | * one for write (we recover only one drive per r10buf) | |
100 | * | |
101 | */ | |
dd0fc66f | 102 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 103 | { |
e879a879 | 104 | struct r10conf *conf = data; |
1da177e4 | 105 | struct page *page; |
9f2c9d12 | 106 | struct r10bio *r10_bio; |
1da177e4 LT |
107 | struct bio *bio; |
108 | int i, j; | |
109 | int nalloc; | |
110 | ||
111 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 112 | if (!r10_bio) |
1da177e4 | 113 | return NULL; |
1da177e4 LT |
114 | |
115 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
116 | nalloc = conf->copies; /* resync */ | |
117 | else | |
118 | nalloc = 2; /* recovery */ | |
119 | ||
120 | /* | |
121 | * Allocate bios. | |
122 | */ | |
123 | for (j = nalloc ; j-- ; ) { | |
6746557f | 124 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
125 | if (!bio) |
126 | goto out_free_bio; | |
127 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
128 | if (!conf->have_replacement) |
129 | continue; | |
130 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
131 | if (!bio) | |
132 | goto out_free_bio; | |
133 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
134 | } |
135 | /* | |
136 | * Allocate RESYNC_PAGES data pages and attach them | |
137 | * where needed. | |
138 | */ | |
139 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 140 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
141 | bio = r10_bio->devs[j].bio; |
142 | for (i = 0; i < RESYNC_PAGES; i++) { | |
c65060ad NK |
143 | if (j == 1 && !test_bit(MD_RECOVERY_SYNC, |
144 | &conf->mddev->recovery)) { | |
145 | /* we can share bv_page's during recovery */ | |
146 | struct bio *rbio = r10_bio->devs[0].bio; | |
147 | page = rbio->bi_io_vec[i].bv_page; | |
148 | get_page(page); | |
149 | } else | |
150 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
151 | if (unlikely(!page)) |
152 | goto out_free_pages; | |
153 | ||
154 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
155 | if (rbio) |
156 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
157 | } |
158 | } | |
159 | ||
160 | return r10_bio; | |
161 | ||
162 | out_free_pages: | |
163 | for ( ; i > 0 ; i--) | |
1345b1d8 | 164 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
165 | while (j--) |
166 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 167 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
168 | j = -1; |
169 | out_free_bio: | |
69335ef3 | 170 | while (++j < nalloc) { |
1da177e4 | 171 | bio_put(r10_bio->devs[j].bio); |
69335ef3 N |
172 | if (r10_bio->devs[j].repl_bio) |
173 | bio_put(r10_bio->devs[j].repl_bio); | |
174 | } | |
1da177e4 LT |
175 | r10bio_pool_free(r10_bio, conf); |
176 | return NULL; | |
177 | } | |
178 | ||
179 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
180 | { | |
181 | int i; | |
e879a879 | 182 | struct r10conf *conf = data; |
9f2c9d12 | 183 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
184 | int j; |
185 | ||
186 | for (j=0; j < conf->copies; j++) { | |
187 | struct bio *bio = r10bio->devs[j].bio; | |
188 | if (bio) { | |
189 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 190 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
191 | bio->bi_io_vec[i].bv_page = NULL; |
192 | } | |
193 | bio_put(bio); | |
194 | } | |
69335ef3 N |
195 | bio = r10bio->devs[j].repl_bio; |
196 | if (bio) | |
197 | bio_put(bio); | |
1da177e4 LT |
198 | } |
199 | r10bio_pool_free(r10bio, conf); | |
200 | } | |
201 | ||
e879a879 | 202 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
203 | { |
204 | int i; | |
205 | ||
206 | for (i = 0; i < conf->copies; i++) { | |
207 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 208 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
209 | bio_put(*bio); |
210 | *bio = NULL; | |
69335ef3 N |
211 | bio = &r10_bio->devs[i].repl_bio; |
212 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
213 | bio_put(*bio); | |
214 | *bio = NULL; | |
1da177e4 LT |
215 | } |
216 | } | |
217 | ||
9f2c9d12 | 218 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 219 | { |
e879a879 | 220 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 221 | |
1da177e4 LT |
222 | put_all_bios(conf, r10_bio); |
223 | mempool_free(r10_bio, conf->r10bio_pool); | |
224 | } | |
225 | ||
9f2c9d12 | 226 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 227 | { |
e879a879 | 228 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
229 | |
230 | mempool_free(r10_bio, conf->r10buf_pool); | |
231 | ||
0a27ec96 | 232 | lower_barrier(conf); |
1da177e4 LT |
233 | } |
234 | ||
9f2c9d12 | 235 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
236 | { |
237 | unsigned long flags; | |
fd01b88c | 238 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 239 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
240 | |
241 | spin_lock_irqsave(&conf->device_lock, flags); | |
242 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 243 | conf->nr_queued ++; |
1da177e4 LT |
244 | spin_unlock_irqrestore(&conf->device_lock, flags); |
245 | ||
388667be AJ |
246 | /* wake up frozen array... */ |
247 | wake_up(&conf->wait_barrier); | |
248 | ||
1da177e4 LT |
249 | md_wakeup_thread(mddev->thread); |
250 | } | |
251 | ||
252 | /* | |
253 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
254 | * operation and are ready to return a success/failure code to the buffer | |
255 | * cache layer. | |
256 | */ | |
9f2c9d12 | 257 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
258 | { |
259 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 260 | int done; |
e879a879 | 261 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 262 | |
856e08e2 N |
263 | if (bio->bi_phys_segments) { |
264 | unsigned long flags; | |
265 | spin_lock_irqsave(&conf->device_lock, flags); | |
266 | bio->bi_phys_segments--; | |
267 | done = (bio->bi_phys_segments == 0); | |
268 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
269 | } else | |
270 | done = 1; | |
271 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
272 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
273 | if (done) { | |
274 | bio_endio(bio, 0); | |
275 | /* | |
276 | * Wake up any possible resync thread that waits for the device | |
277 | * to go idle. | |
278 | */ | |
279 | allow_barrier(conf); | |
280 | } | |
1da177e4 LT |
281 | free_r10bio(r10_bio); |
282 | } | |
283 | ||
284 | /* | |
285 | * Update disk head position estimator based on IRQ completion info. | |
286 | */ | |
9f2c9d12 | 287 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 288 | { |
e879a879 | 289 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
290 | |
291 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
292 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
293 | } | |
294 | ||
778ca018 NK |
295 | /* |
296 | * Find the disk number which triggered given bio | |
297 | */ | |
e879a879 | 298 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 299 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
300 | { |
301 | int slot; | |
69335ef3 | 302 | int repl = 0; |
778ca018 | 303 | |
69335ef3 | 304 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
305 | if (r10_bio->devs[slot].bio == bio) |
306 | break; | |
69335ef3 N |
307 | if (r10_bio->devs[slot].repl_bio == bio) { |
308 | repl = 1; | |
309 | break; | |
310 | } | |
311 | } | |
778ca018 NK |
312 | |
313 | BUG_ON(slot == conf->copies); | |
314 | update_head_pos(slot, r10_bio); | |
315 | ||
749c55e9 N |
316 | if (slotp) |
317 | *slotp = slot; | |
69335ef3 N |
318 | if (replp) |
319 | *replp = repl; | |
778ca018 NK |
320 | return r10_bio->devs[slot].devnum; |
321 | } | |
322 | ||
6712ecf8 | 323 | static void raid10_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
324 | { |
325 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 326 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 327 | int slot, dev; |
abbf098e | 328 | struct md_rdev *rdev; |
e879a879 | 329 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 330 | |
1da177e4 LT |
331 | |
332 | slot = r10_bio->read_slot; | |
333 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 334 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
335 | /* |
336 | * this branch is our 'one mirror IO has finished' event handler: | |
337 | */ | |
4443ae10 N |
338 | update_head_pos(slot, r10_bio); |
339 | ||
340 | if (uptodate) { | |
1da177e4 LT |
341 | /* |
342 | * Set R10BIO_Uptodate in our master bio, so that | |
343 | * we will return a good error code to the higher | |
344 | * levels even if IO on some other mirrored buffer fails. | |
345 | * | |
346 | * The 'master' represents the composite IO operation to | |
347 | * user-side. So if something waits for IO, then it will | |
348 | * wait for the 'master' bio. | |
349 | */ | |
350 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
351 | } else { |
352 | /* If all other devices that store this block have | |
353 | * failed, we want to return the error upwards rather | |
354 | * than fail the last device. Here we redefine | |
355 | * "uptodate" to mean "Don't want to retry" | |
356 | */ | |
357 | unsigned long flags; | |
358 | spin_lock_irqsave(&conf->device_lock, flags); | |
359 | if (!enough(conf, rdev->raid_disk)) | |
360 | uptodate = 1; | |
361 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
362 | } | |
363 | if (uptodate) { | |
1da177e4 | 364 | raid_end_bio_io(r10_bio); |
abbf098e | 365 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 366 | } else { |
1da177e4 | 367 | /* |
7c4e06ff | 368 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
369 | */ |
370 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
371 | printk_ratelimited(KERN_ERR |
372 | "md/raid10:%s: %s: rescheduling sector %llu\n", | |
373 | mdname(conf->mddev), | |
abbf098e | 374 | bdevname(rdev->bdev, b), |
8bda470e | 375 | (unsigned long long)r10_bio->sector); |
856e08e2 | 376 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
377 | reschedule_retry(r10_bio); |
378 | } | |
1da177e4 LT |
379 | } |
380 | ||
9f2c9d12 | 381 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
382 | { |
383 | /* clear the bitmap if all writes complete successfully */ | |
384 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
385 | r10_bio->sectors, | |
386 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
387 | 0); | |
388 | md_write_end(r10_bio->mddev); | |
389 | } | |
390 | ||
9f2c9d12 | 391 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
392 | { |
393 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
394 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
395 | reschedule_retry(r10_bio); | |
396 | else { | |
397 | close_write(r10_bio); | |
398 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
399 | reschedule_retry(r10_bio); | |
400 | else | |
401 | raid_end_bio_io(r10_bio); | |
402 | } | |
403 | } | |
404 | } | |
405 | ||
6712ecf8 | 406 | static void raid10_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
407 | { |
408 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 409 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 410 | int dev; |
749c55e9 | 411 | int dec_rdev = 1; |
e879a879 | 412 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 413 | int slot, repl; |
4ca40c2c | 414 | struct md_rdev *rdev = NULL; |
1da177e4 | 415 | |
475b0321 | 416 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 417 | |
475b0321 N |
418 | if (repl) |
419 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
420 | if (!rdev) { |
421 | smp_rmb(); | |
422 | repl = 0; | |
475b0321 | 423 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 424 | } |
1da177e4 LT |
425 | /* |
426 | * this branch is our 'one mirror IO has finished' event handler: | |
427 | */ | |
6cce3b23 | 428 | if (!uptodate) { |
475b0321 N |
429 | if (repl) |
430 | /* Never record new bad blocks to replacement, | |
431 | * just fail it. | |
432 | */ | |
433 | md_error(rdev->mddev, rdev); | |
434 | else { | |
435 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
436 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
437 | set_bit(MD_RECOVERY_NEEDED, | |
438 | &rdev->mddev->recovery); | |
475b0321 N |
439 | set_bit(R10BIO_WriteError, &r10_bio->state); |
440 | dec_rdev = 0; | |
441 | } | |
749c55e9 | 442 | } else { |
1da177e4 LT |
443 | /* |
444 | * Set R10BIO_Uptodate in our master bio, so that | |
445 | * we will return a good error code for to the higher | |
446 | * levels even if IO on some other mirrored buffer fails. | |
447 | * | |
448 | * The 'master' represents the composite IO operation to | |
449 | * user-side. So if something waits for IO, then it will | |
450 | * wait for the 'master' bio. | |
451 | */ | |
749c55e9 N |
452 | sector_t first_bad; |
453 | int bad_sectors; | |
454 | ||
1da177e4 LT |
455 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
456 | ||
749c55e9 | 457 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 458 | if (is_badblock(rdev, |
749c55e9 N |
459 | r10_bio->devs[slot].addr, |
460 | r10_bio->sectors, | |
461 | &first_bad, &bad_sectors)) { | |
462 | bio_put(bio); | |
475b0321 N |
463 | if (repl) |
464 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
465 | else | |
466 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
467 | dec_rdev = 0; |
468 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
469 | } | |
470 | } | |
471 | ||
1da177e4 LT |
472 | /* |
473 | * | |
474 | * Let's see if all mirrored write operations have finished | |
475 | * already. | |
476 | */ | |
19d5f834 | 477 | one_write_done(r10_bio); |
749c55e9 N |
478 | if (dec_rdev) |
479 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
480 | } |
481 | ||
1da177e4 LT |
482 | /* |
483 | * RAID10 layout manager | |
25985edc | 484 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
485 | * parameters: near_copies and far_copies. |
486 | * near_copies * far_copies must be <= raid_disks. | |
487 | * Normally one of these will be 1. | |
488 | * If both are 1, we get raid0. | |
489 | * If near_copies == raid_disks, we get raid1. | |
490 | * | |
25985edc | 491 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
492 | * first chunk, followed by near_copies copies of the next chunk and |
493 | * so on. | |
494 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
495 | * as described above, we start again with a device offset of near_copies. | |
496 | * So we effectively have another copy of the whole array further down all | |
497 | * the drives, but with blocks on different drives. | |
498 | * With this layout, and block is never stored twice on the one device. | |
499 | * | |
500 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 501 | * on each device that it is on. |
1da177e4 LT |
502 | * |
503 | * raid10_find_virt does the reverse mapping, from a device and a | |
504 | * sector offset to a virtual address | |
505 | */ | |
506 | ||
e879a879 | 507 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) |
1da177e4 LT |
508 | { |
509 | int n,f; | |
510 | sector_t sector; | |
511 | sector_t chunk; | |
512 | sector_t stripe; | |
513 | int dev; | |
514 | ||
515 | int slot = 0; | |
516 | ||
517 | /* now calculate first sector/dev */ | |
518 | chunk = r10bio->sector >> conf->chunk_shift; | |
519 | sector = r10bio->sector & conf->chunk_mask; | |
520 | ||
521 | chunk *= conf->near_copies; | |
522 | stripe = chunk; | |
523 | dev = sector_div(stripe, conf->raid_disks); | |
c93983bf N |
524 | if (conf->far_offset) |
525 | stripe *= conf->far_copies; | |
1da177e4 LT |
526 | |
527 | sector += stripe << conf->chunk_shift; | |
528 | ||
529 | /* and calculate all the others */ | |
530 | for (n=0; n < conf->near_copies; n++) { | |
531 | int d = dev; | |
532 | sector_t s = sector; | |
533 | r10bio->devs[slot].addr = sector; | |
534 | r10bio->devs[slot].devnum = d; | |
535 | slot++; | |
536 | ||
537 | for (f = 1; f < conf->far_copies; f++) { | |
538 | d += conf->near_copies; | |
539 | if (d >= conf->raid_disks) | |
540 | d -= conf->raid_disks; | |
541 | s += conf->stride; | |
542 | r10bio->devs[slot].devnum = d; | |
543 | r10bio->devs[slot].addr = s; | |
544 | slot++; | |
545 | } | |
546 | dev++; | |
547 | if (dev >= conf->raid_disks) { | |
548 | dev = 0; | |
549 | sector += (conf->chunk_mask + 1); | |
550 | } | |
551 | } | |
552 | BUG_ON(slot != conf->copies); | |
553 | } | |
554 | ||
e879a879 | 555 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
556 | { |
557 | sector_t offset, chunk, vchunk; | |
558 | ||
1da177e4 | 559 | offset = sector & conf->chunk_mask; |
c93983bf N |
560 | if (conf->far_offset) { |
561 | int fc; | |
562 | chunk = sector >> conf->chunk_shift; | |
563 | fc = sector_div(chunk, conf->far_copies); | |
564 | dev -= fc * conf->near_copies; | |
565 | if (dev < 0) | |
566 | dev += conf->raid_disks; | |
567 | } else { | |
64a742bc | 568 | while (sector >= conf->stride) { |
c93983bf N |
569 | sector -= conf->stride; |
570 | if (dev < conf->near_copies) | |
571 | dev += conf->raid_disks - conf->near_copies; | |
572 | else | |
573 | dev -= conf->near_copies; | |
574 | } | |
575 | chunk = sector >> conf->chunk_shift; | |
576 | } | |
1da177e4 LT |
577 | vchunk = chunk * conf->raid_disks + dev; |
578 | sector_div(vchunk, conf->near_copies); | |
579 | return (vchunk << conf->chunk_shift) + offset; | |
580 | } | |
581 | ||
582 | /** | |
583 | * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged | |
584 | * @q: request queue | |
cc371e66 | 585 | * @bvm: properties of new bio |
1da177e4 LT |
586 | * @biovec: the request that could be merged to it. |
587 | * | |
588 | * Return amount of bytes we can accept at this offset | |
050b6615 N |
589 | * This requires checking for end-of-chunk if near_copies != raid_disks, |
590 | * and for subordinate merge_bvec_fns if merge_check_needed. | |
1da177e4 | 591 | */ |
cc371e66 AK |
592 | static int raid10_mergeable_bvec(struct request_queue *q, |
593 | struct bvec_merge_data *bvm, | |
594 | struct bio_vec *biovec) | |
1da177e4 | 595 | { |
fd01b88c | 596 | struct mddev *mddev = q->queuedata; |
050b6615 | 597 | struct r10conf *conf = mddev->private; |
cc371e66 | 598 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 599 | int max; |
9d8f0363 | 600 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 601 | unsigned int bio_sectors = bvm->bi_size >> 9; |
1da177e4 | 602 | |
050b6615 N |
603 | if (conf->near_copies < conf->raid_disks) { |
604 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) | |
605 | + bio_sectors)) << 9; | |
606 | if (max < 0) | |
607 | /* bio_add cannot handle a negative return */ | |
608 | max = 0; | |
609 | if (max <= biovec->bv_len && bio_sectors == 0) | |
610 | return biovec->bv_len; | |
611 | } else | |
612 | max = biovec->bv_len; | |
613 | ||
614 | if (mddev->merge_check_needed) { | |
615 | struct r10bio r10_bio; | |
616 | int s; | |
617 | r10_bio.sector = sector; | |
618 | raid10_find_phys(conf, &r10_bio); | |
619 | rcu_read_lock(); | |
620 | for (s = 0; s < conf->copies; s++) { | |
621 | int disk = r10_bio.devs[s].devnum; | |
622 | struct md_rdev *rdev = rcu_dereference( | |
623 | conf->mirrors[disk].rdev); | |
624 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
625 | struct request_queue *q = | |
626 | bdev_get_queue(rdev->bdev); | |
627 | if (q->merge_bvec_fn) { | |
628 | bvm->bi_sector = r10_bio.devs[s].addr | |
629 | + rdev->data_offset; | |
630 | bvm->bi_bdev = rdev->bdev; | |
631 | max = min(max, q->merge_bvec_fn( | |
632 | q, bvm, biovec)); | |
633 | } | |
634 | } | |
635 | rdev = rcu_dereference(conf->mirrors[disk].replacement); | |
636 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
637 | struct request_queue *q = | |
638 | bdev_get_queue(rdev->bdev); | |
639 | if (q->merge_bvec_fn) { | |
640 | bvm->bi_sector = r10_bio.devs[s].addr | |
641 | + rdev->data_offset; | |
642 | bvm->bi_bdev = rdev->bdev; | |
643 | max = min(max, q->merge_bvec_fn( | |
644 | q, bvm, biovec)); | |
645 | } | |
646 | } | |
647 | } | |
648 | rcu_read_unlock(); | |
649 | } | |
650 | return max; | |
1da177e4 LT |
651 | } |
652 | ||
653 | /* | |
654 | * This routine returns the disk from which the requested read should | |
655 | * be done. There is a per-array 'next expected sequential IO' sector | |
656 | * number - if this matches on the next IO then we use the last disk. | |
657 | * There is also a per-disk 'last know head position' sector that is | |
658 | * maintained from IRQ contexts, both the normal and the resync IO | |
659 | * completion handlers update this position correctly. If there is no | |
660 | * perfect sequential match then we pick the disk whose head is closest. | |
661 | * | |
662 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
663 | * because position is mirror, not device based. | |
664 | * | |
665 | * The rdev for the device selected will have nr_pending incremented. | |
666 | */ | |
667 | ||
668 | /* | |
669 | * FIXME: possibly should rethink readbalancing and do it differently | |
670 | * depending on near_copies / far_copies geometry. | |
671 | */ | |
96c3fd1f N |
672 | static struct md_rdev *read_balance(struct r10conf *conf, |
673 | struct r10bio *r10_bio, | |
674 | int *max_sectors) | |
1da177e4 | 675 | { |
af3a2cd6 | 676 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 677 | int disk, slot; |
856e08e2 N |
678 | int sectors = r10_bio->sectors; |
679 | int best_good_sectors; | |
56d99121 | 680 | sector_t new_distance, best_dist; |
abbf098e | 681 | struct md_rdev *rdev, *best_rdev; |
56d99121 N |
682 | int do_balance; |
683 | int best_slot; | |
1da177e4 LT |
684 | |
685 | raid10_find_phys(conf, r10_bio); | |
686 | rcu_read_lock(); | |
56d99121 | 687 | retry: |
856e08e2 | 688 | sectors = r10_bio->sectors; |
56d99121 | 689 | best_slot = -1; |
abbf098e | 690 | best_rdev = NULL; |
56d99121 | 691 | best_dist = MaxSector; |
856e08e2 | 692 | best_good_sectors = 0; |
56d99121 | 693 | do_balance = 1; |
1da177e4 LT |
694 | /* |
695 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
696 | * device if no resync is going on (recovery is ok), or below |
697 | * the resync window. We take the first readable disk when | |
698 | * above the resync window. | |
1da177e4 LT |
699 | */ |
700 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
701 | && (this_sector + sectors >= conf->next_resync)) |
702 | do_balance = 0; | |
1da177e4 | 703 | |
56d99121 | 704 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
705 | sector_t first_bad; |
706 | int bad_sectors; | |
707 | sector_t dev_sector; | |
708 | ||
56d99121 N |
709 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
710 | continue; | |
1da177e4 | 711 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
712 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
713 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
050b6615 | 714 | test_bit(Unmerged, &rdev->flags) || |
abbf098e N |
715 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) |
716 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 N |
717 | if (rdev == NULL || |
718 | test_bit(Faulty, &rdev->flags) || | |
719 | test_bit(Unmerged, &rdev->flags)) | |
abbf098e N |
720 | continue; |
721 | if (!test_bit(In_sync, &rdev->flags) && | |
722 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
723 | continue; |
724 | ||
856e08e2 N |
725 | dev_sector = r10_bio->devs[slot].addr; |
726 | if (is_badblock(rdev, dev_sector, sectors, | |
727 | &first_bad, &bad_sectors)) { | |
728 | if (best_dist < MaxSector) | |
729 | /* Already have a better slot */ | |
730 | continue; | |
731 | if (first_bad <= dev_sector) { | |
732 | /* Cannot read here. If this is the | |
733 | * 'primary' device, then we must not read | |
734 | * beyond 'bad_sectors' from another device. | |
735 | */ | |
736 | bad_sectors -= (dev_sector - first_bad); | |
737 | if (!do_balance && sectors > bad_sectors) | |
738 | sectors = bad_sectors; | |
739 | if (best_good_sectors > sectors) | |
740 | best_good_sectors = sectors; | |
741 | } else { | |
742 | sector_t good_sectors = | |
743 | first_bad - dev_sector; | |
744 | if (good_sectors > best_good_sectors) { | |
745 | best_good_sectors = good_sectors; | |
746 | best_slot = slot; | |
abbf098e | 747 | best_rdev = rdev; |
856e08e2 N |
748 | } |
749 | if (!do_balance) | |
750 | /* Must read from here */ | |
751 | break; | |
752 | } | |
753 | continue; | |
754 | } else | |
755 | best_good_sectors = sectors; | |
756 | ||
56d99121 N |
757 | if (!do_balance) |
758 | break; | |
1da177e4 | 759 | |
22dfdf52 N |
760 | /* This optimisation is debatable, and completely destroys |
761 | * sequential read speed for 'far copies' arrays. So only | |
762 | * keep it for 'near' arrays, and review those later. | |
763 | */ | |
56d99121 | 764 | if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
1da177e4 | 765 | break; |
8ed3a195 KS |
766 | |
767 | /* for far > 1 always use the lowest address */ | |
768 | if (conf->far_copies > 1) | |
56d99121 | 769 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 770 | else |
56d99121 N |
771 | new_distance = abs(r10_bio->devs[slot].addr - |
772 | conf->mirrors[disk].head_position); | |
773 | if (new_distance < best_dist) { | |
774 | best_dist = new_distance; | |
775 | best_slot = slot; | |
abbf098e | 776 | best_rdev = rdev; |
1da177e4 LT |
777 | } |
778 | } | |
abbf098e | 779 | if (slot >= conf->copies) { |
56d99121 | 780 | slot = best_slot; |
abbf098e N |
781 | rdev = best_rdev; |
782 | } | |
1da177e4 | 783 | |
56d99121 | 784 | if (slot >= 0) { |
56d99121 N |
785 | atomic_inc(&rdev->nr_pending); |
786 | if (test_bit(Faulty, &rdev->flags)) { | |
787 | /* Cannot risk returning a device that failed | |
788 | * before we inc'ed nr_pending | |
789 | */ | |
790 | rdev_dec_pending(rdev, conf->mddev); | |
791 | goto retry; | |
792 | } | |
793 | r10_bio->read_slot = slot; | |
794 | } else | |
96c3fd1f | 795 | rdev = NULL; |
1da177e4 | 796 | rcu_read_unlock(); |
856e08e2 | 797 | *max_sectors = best_good_sectors; |
1da177e4 | 798 | |
96c3fd1f | 799 | return rdev; |
1da177e4 LT |
800 | } |
801 | ||
0d129228 N |
802 | static int raid10_congested(void *data, int bits) |
803 | { | |
fd01b88c | 804 | struct mddev *mddev = data; |
e879a879 | 805 | struct r10conf *conf = mddev->private; |
0d129228 N |
806 | int i, ret = 0; |
807 | ||
34db0cd6 N |
808 | if ((bits & (1 << BDI_async_congested)) && |
809 | conf->pending_count >= max_queued_requests) | |
810 | return 1; | |
811 | ||
3fa841d7 N |
812 | if (mddev_congested(mddev, bits)) |
813 | return 1; | |
0d129228 | 814 | rcu_read_lock(); |
84707f38 | 815 | for (i = 0; i < conf->raid_disks && ret == 0; i++) { |
3cb03002 | 816 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 817 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 818 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
819 | |
820 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
821 | } | |
822 | } | |
823 | rcu_read_unlock(); | |
824 | return ret; | |
825 | } | |
826 | ||
e879a879 | 827 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
828 | { |
829 | /* Any writes that have been queued but are awaiting | |
830 | * bitmap updates get flushed here. | |
a35e63ef | 831 | */ |
a35e63ef N |
832 | spin_lock_irq(&conf->device_lock); |
833 | ||
834 | if (conf->pending_bio_list.head) { | |
835 | struct bio *bio; | |
836 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 837 | conf->pending_count = 0; |
a35e63ef N |
838 | spin_unlock_irq(&conf->device_lock); |
839 | /* flush any pending bitmap writes to disk | |
840 | * before proceeding w/ I/O */ | |
841 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 842 | wake_up(&conf->wait_barrier); |
a35e63ef N |
843 | |
844 | while (bio) { /* submit pending writes */ | |
845 | struct bio *next = bio->bi_next; | |
846 | bio->bi_next = NULL; | |
847 | generic_make_request(bio); | |
848 | bio = next; | |
849 | } | |
a35e63ef N |
850 | } else |
851 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 852 | } |
7eaceacc | 853 | |
0a27ec96 N |
854 | /* Barriers.... |
855 | * Sometimes we need to suspend IO while we do something else, | |
856 | * either some resync/recovery, or reconfigure the array. | |
857 | * To do this we raise a 'barrier'. | |
858 | * The 'barrier' is a counter that can be raised multiple times | |
859 | * to count how many activities are happening which preclude | |
860 | * normal IO. | |
861 | * We can only raise the barrier if there is no pending IO. | |
862 | * i.e. if nr_pending == 0. | |
863 | * We choose only to raise the barrier if no-one is waiting for the | |
864 | * barrier to go down. This means that as soon as an IO request | |
865 | * is ready, no other operations which require a barrier will start | |
866 | * until the IO request has had a chance. | |
867 | * | |
868 | * So: regular IO calls 'wait_barrier'. When that returns there | |
869 | * is no backgroup IO happening, It must arrange to call | |
870 | * allow_barrier when it has finished its IO. | |
871 | * backgroup IO calls must call raise_barrier. Once that returns | |
872 | * there is no normal IO happeing. It must arrange to call | |
873 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 874 | */ |
1da177e4 | 875 | |
e879a879 | 876 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 877 | { |
6cce3b23 | 878 | BUG_ON(force && !conf->barrier); |
1da177e4 | 879 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 880 | |
6cce3b23 N |
881 | /* Wait until no block IO is waiting (unless 'force') */ |
882 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
c3b328ac | 883 | conf->resync_lock, ); |
0a27ec96 N |
884 | |
885 | /* block any new IO from starting */ | |
886 | conf->barrier++; | |
887 | ||
c3b328ac | 888 | /* Now wait for all pending IO to complete */ |
0a27ec96 N |
889 | wait_event_lock_irq(conf->wait_barrier, |
890 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
c3b328ac | 891 | conf->resync_lock, ); |
0a27ec96 N |
892 | |
893 | spin_unlock_irq(&conf->resync_lock); | |
894 | } | |
895 | ||
e879a879 | 896 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
897 | { |
898 | unsigned long flags; | |
899 | spin_lock_irqsave(&conf->resync_lock, flags); | |
900 | conf->barrier--; | |
901 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
902 | wake_up(&conf->wait_barrier); | |
903 | } | |
904 | ||
e879a879 | 905 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
906 | { |
907 | spin_lock_irq(&conf->resync_lock); | |
908 | if (conf->barrier) { | |
909 | conf->nr_waiting++; | |
d6b42dcb N |
910 | /* Wait for the barrier to drop. |
911 | * However if there are already pending | |
912 | * requests (preventing the barrier from | |
913 | * rising completely), and the | |
914 | * pre-process bio queue isn't empty, | |
915 | * then don't wait, as we need to empty | |
916 | * that queue to get the nr_pending | |
917 | * count down. | |
918 | */ | |
919 | wait_event_lock_irq(conf->wait_barrier, | |
920 | !conf->barrier || | |
921 | (conf->nr_pending && | |
922 | current->bio_list && | |
923 | !bio_list_empty(current->bio_list)), | |
0a27ec96 | 924 | conf->resync_lock, |
d6b42dcb | 925 | ); |
0a27ec96 | 926 | conf->nr_waiting--; |
1da177e4 | 927 | } |
0a27ec96 | 928 | conf->nr_pending++; |
1da177e4 LT |
929 | spin_unlock_irq(&conf->resync_lock); |
930 | } | |
931 | ||
e879a879 | 932 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 N |
933 | { |
934 | unsigned long flags; | |
935 | spin_lock_irqsave(&conf->resync_lock, flags); | |
936 | conf->nr_pending--; | |
937 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
938 | wake_up(&conf->wait_barrier); | |
939 | } | |
940 | ||
e879a879 | 941 | static void freeze_array(struct r10conf *conf) |
4443ae10 N |
942 | { |
943 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 944 | * go quiet. |
4443ae10 | 945 | * We increment barrier and nr_waiting, and then |
1c830532 N |
946 | * wait until nr_pending match nr_queued+1 |
947 | * This is called in the context of one normal IO request | |
948 | * that has failed. Thus any sync request that might be pending | |
949 | * will be blocked by nr_pending, and we need to wait for | |
950 | * pending IO requests to complete or be queued for re-try. | |
951 | * Thus the number queued (nr_queued) plus this request (1) | |
952 | * must match the number of pending IOs (nr_pending) before | |
953 | * we continue. | |
4443ae10 N |
954 | */ |
955 | spin_lock_irq(&conf->resync_lock); | |
956 | conf->barrier++; | |
957 | conf->nr_waiting++; | |
958 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 959 | conf->nr_pending == conf->nr_queued+1, |
4443ae10 | 960 | conf->resync_lock, |
c3b328ac N |
961 | flush_pending_writes(conf)); |
962 | ||
4443ae10 N |
963 | spin_unlock_irq(&conf->resync_lock); |
964 | } | |
965 | ||
e879a879 | 966 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
967 | { |
968 | /* reverse the effect of the freeze */ | |
969 | spin_lock_irq(&conf->resync_lock); | |
970 | conf->barrier--; | |
971 | conf->nr_waiting--; | |
972 | wake_up(&conf->wait_barrier); | |
973 | spin_unlock_irq(&conf->resync_lock); | |
974 | } | |
975 | ||
b4fdcb02 | 976 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 977 | { |
e879a879 | 978 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 979 | struct r10bio *r10_bio; |
1da177e4 LT |
980 | struct bio *read_bio; |
981 | int i; | |
982 | int chunk_sects = conf->chunk_mask + 1; | |
a362357b | 983 | const int rw = bio_data_dir(bio); |
2c7d46ec | 984 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 985 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
6cce3b23 | 986 | unsigned long flags; |
3cb03002 | 987 | struct md_rdev *blocked_rdev; |
c3b328ac | 988 | int plugged; |
d4432c23 N |
989 | int sectors_handled; |
990 | int max_sectors; | |
1da177e4 | 991 | |
e9c7469b TH |
992 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
993 | md_flush_request(mddev, bio); | |
5a7bbad2 | 994 | return; |
e5dcdd80 N |
995 | } |
996 | ||
1da177e4 LT |
997 | /* If this request crosses a chunk boundary, we need to |
998 | * split it. This will only happen for 1 PAGE (or less) requests. | |
999 | */ | |
1000 | if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9) | |
1001 | > chunk_sects && | |
1002 | conf->near_copies < conf->raid_disks)) { | |
1003 | struct bio_pair *bp; | |
1004 | /* Sanity check -- queue functions should prevent this happening */ | |
1005 | if (bio->bi_vcnt != 1 || | |
1006 | bio->bi_idx != 0) | |
1007 | goto bad_map; | |
1008 | /* This is a one page bio that upper layers | |
1009 | * refuse to split for us, so we need to split it. | |
1010 | */ | |
6feef531 | 1011 | bp = bio_split(bio, |
1da177e4 | 1012 | chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); |
51e9ac77 N |
1013 | |
1014 | /* Each of these 'make_request' calls will call 'wait_barrier'. | |
1015 | * If the first succeeds but the second blocks due to the resync | |
1016 | * thread raising the barrier, we will deadlock because the | |
1017 | * IO to the underlying device will be queued in generic_make_request | |
1018 | * and will never complete, so will never reduce nr_pending. | |
1019 | * So increment nr_waiting here so no new raise_barriers will | |
1020 | * succeed, and so the second wait_barrier cannot block. | |
1021 | */ | |
1022 | spin_lock_irq(&conf->resync_lock); | |
1023 | conf->nr_waiting++; | |
1024 | spin_unlock_irq(&conf->resync_lock); | |
1025 | ||
5a7bbad2 CH |
1026 | make_request(mddev, &bp->bio1); |
1027 | make_request(mddev, &bp->bio2); | |
1da177e4 | 1028 | |
51e9ac77 N |
1029 | spin_lock_irq(&conf->resync_lock); |
1030 | conf->nr_waiting--; | |
1031 | wake_up(&conf->wait_barrier); | |
1032 | spin_unlock_irq(&conf->resync_lock); | |
1033 | ||
1da177e4 | 1034 | bio_pair_release(bp); |
5a7bbad2 | 1035 | return; |
1da177e4 | 1036 | bad_map: |
128595ed N |
1037 | printk("md/raid10:%s: make_request bug: can't convert block across chunks" |
1038 | " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2, | |
1da177e4 LT |
1039 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); |
1040 | ||
6712ecf8 | 1041 | bio_io_error(bio); |
5a7bbad2 | 1042 | return; |
1da177e4 LT |
1043 | } |
1044 | ||
3d310eb7 | 1045 | md_write_start(mddev, bio); |
06d91a5f | 1046 | |
1da177e4 LT |
1047 | /* |
1048 | * Register the new request and wait if the reconstruction | |
1049 | * thread has put up a bar for new requests. | |
1050 | * Continue immediately if no resync is active currently. | |
1051 | */ | |
0a27ec96 | 1052 | wait_barrier(conf); |
1da177e4 | 1053 | |
1da177e4 LT |
1054 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
1055 | ||
1056 | r10_bio->master_bio = bio; | |
1057 | r10_bio->sectors = bio->bi_size >> 9; | |
1058 | ||
1059 | r10_bio->mddev = mddev; | |
1060 | r10_bio->sector = bio->bi_sector; | |
6cce3b23 | 1061 | r10_bio->state = 0; |
1da177e4 | 1062 | |
856e08e2 N |
1063 | /* We might need to issue multiple reads to different |
1064 | * devices if there are bad blocks around, so we keep | |
1065 | * track of the number of reads in bio->bi_phys_segments. | |
1066 | * If this is 0, there is only one r10_bio and no locking | |
1067 | * will be needed when the request completes. If it is | |
1068 | * non-zero, then it is the number of not-completed requests. | |
1069 | */ | |
1070 | bio->bi_phys_segments = 0; | |
1071 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1072 | ||
a362357b | 1073 | if (rw == READ) { |
1da177e4 LT |
1074 | /* |
1075 | * read balancing logic: | |
1076 | */ | |
96c3fd1f | 1077 | struct md_rdev *rdev; |
856e08e2 N |
1078 | int slot; |
1079 | ||
1080 | read_again: | |
96c3fd1f N |
1081 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1082 | if (!rdev) { | |
1da177e4 | 1083 | raid_end_bio_io(r10_bio); |
5a7bbad2 | 1084 | return; |
1da177e4 | 1085 | } |
96c3fd1f | 1086 | slot = r10_bio->read_slot; |
1da177e4 | 1087 | |
a167f663 | 1088 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
856e08e2 N |
1089 | md_trim_bio(read_bio, r10_bio->sector - bio->bi_sector, |
1090 | max_sectors); | |
1da177e4 LT |
1091 | |
1092 | r10_bio->devs[slot].bio = read_bio; | |
abbf098e | 1093 | r10_bio->devs[slot].rdev = rdev; |
1da177e4 LT |
1094 | |
1095 | read_bio->bi_sector = r10_bio->devs[slot].addr + | |
96c3fd1f N |
1096 | rdev->data_offset; |
1097 | read_bio->bi_bdev = rdev->bdev; | |
1da177e4 | 1098 | read_bio->bi_end_io = raid10_end_read_request; |
7b6d91da | 1099 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1100 | read_bio->bi_private = r10_bio; |
1101 | ||
856e08e2 N |
1102 | if (max_sectors < r10_bio->sectors) { |
1103 | /* Could not read all from this device, so we will | |
1104 | * need another r10_bio. | |
1105 | */ | |
856e08e2 N |
1106 | sectors_handled = (r10_bio->sectors + max_sectors |
1107 | - bio->bi_sector); | |
1108 | r10_bio->sectors = max_sectors; | |
1109 | spin_lock_irq(&conf->device_lock); | |
1110 | if (bio->bi_phys_segments == 0) | |
1111 | bio->bi_phys_segments = 2; | |
1112 | else | |
1113 | bio->bi_phys_segments++; | |
1114 | spin_unlock(&conf->device_lock); | |
1115 | /* Cannot call generic_make_request directly | |
1116 | * as that will be queued in __generic_make_request | |
1117 | * and subsequent mempool_alloc might block | |
1118 | * waiting for it. so hand bio over to raid10d. | |
1119 | */ | |
1120 | reschedule_retry(r10_bio); | |
1121 | ||
1122 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1123 | ||
1124 | r10_bio->master_bio = bio; | |
1125 | r10_bio->sectors = ((bio->bi_size >> 9) | |
1126 | - sectors_handled); | |
1127 | r10_bio->state = 0; | |
1128 | r10_bio->mddev = mddev; | |
1129 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1130 | goto read_again; | |
1131 | } else | |
1132 | generic_make_request(read_bio); | |
5a7bbad2 | 1133 | return; |
1da177e4 LT |
1134 | } |
1135 | ||
1136 | /* | |
1137 | * WRITE: | |
1138 | */ | |
34db0cd6 N |
1139 | if (conf->pending_count >= max_queued_requests) { |
1140 | md_wakeup_thread(mddev->thread); | |
1141 | wait_event(conf->wait_barrier, | |
1142 | conf->pending_count < max_queued_requests); | |
1143 | } | |
6bfe0b49 | 1144 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1145 | * inc refcount on their rdev. Record them by setting |
1146 | * bios[x] to bio | |
d4432c23 N |
1147 | * If there are known/acknowledged bad blocks on any device |
1148 | * on which we have seen a write error, we want to avoid | |
1149 | * writing to those blocks. This potentially requires several | |
1150 | * writes to write around the bad blocks. Each set of writes | |
1151 | * gets its own r10_bio with a set of bios attached. The number | |
1152 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1153 | * the read case. | |
1da177e4 | 1154 | */ |
c3b328ac N |
1155 | plugged = mddev_check_plugged(mddev); |
1156 | ||
69335ef3 | 1157 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1158 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1159 | retry_write: |
cb6969e8 | 1160 | blocked_rdev = NULL; |
1da177e4 | 1161 | rcu_read_lock(); |
d4432c23 N |
1162 | max_sectors = r10_bio->sectors; |
1163 | ||
1da177e4 LT |
1164 | for (i = 0; i < conf->copies; i++) { |
1165 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1166 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1167 | struct md_rdev *rrdev = rcu_dereference( |
1168 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1169 | if (rdev == rrdev) |
1170 | rrdev = NULL; | |
6bfe0b49 DW |
1171 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1172 | atomic_inc(&rdev->nr_pending); | |
1173 | blocked_rdev = rdev; | |
1174 | break; | |
1175 | } | |
475b0321 N |
1176 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1177 | atomic_inc(&rrdev->nr_pending); | |
1178 | blocked_rdev = rrdev; | |
1179 | break; | |
1180 | } | |
050b6615 N |
1181 | if (rrdev && (test_bit(Faulty, &rrdev->flags) |
1182 | || test_bit(Unmerged, &rrdev->flags))) | |
475b0321 N |
1183 | rrdev = NULL; |
1184 | ||
d4432c23 | 1185 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1186 | r10_bio->devs[i].repl_bio = NULL; |
050b6615 N |
1187 | if (!rdev || test_bit(Faulty, &rdev->flags) || |
1188 | test_bit(Unmerged, &rdev->flags)) { | |
6cce3b23 | 1189 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1190 | continue; |
1191 | } | |
1192 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1193 | sector_t first_bad; | |
1194 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1195 | int bad_sectors; | |
1196 | int is_bad; | |
1197 | ||
1198 | is_bad = is_badblock(rdev, dev_sector, | |
1199 | max_sectors, | |
1200 | &first_bad, &bad_sectors); | |
1201 | if (is_bad < 0) { | |
1202 | /* Mustn't write here until the bad block | |
1203 | * is acknowledged | |
1204 | */ | |
1205 | atomic_inc(&rdev->nr_pending); | |
1206 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1207 | blocked_rdev = rdev; | |
1208 | break; | |
1209 | } | |
1210 | if (is_bad && first_bad <= dev_sector) { | |
1211 | /* Cannot write here at all */ | |
1212 | bad_sectors -= (dev_sector - first_bad); | |
1213 | if (bad_sectors < max_sectors) | |
1214 | /* Mustn't write more than bad_sectors | |
1215 | * to other devices yet | |
1216 | */ | |
1217 | max_sectors = bad_sectors; | |
1218 | /* We don't set R10BIO_Degraded as that | |
1219 | * only applies if the disk is missing, | |
1220 | * so it might be re-added, and we want to | |
1221 | * know to recover this chunk. | |
1222 | * In this case the device is here, and the | |
1223 | * fact that this chunk is not in-sync is | |
1224 | * recorded in the bad block log. | |
1225 | */ | |
1226 | continue; | |
1227 | } | |
1228 | if (is_bad) { | |
1229 | int good_sectors = first_bad - dev_sector; | |
1230 | if (good_sectors < max_sectors) | |
1231 | max_sectors = good_sectors; | |
1232 | } | |
6cce3b23 | 1233 | } |
d4432c23 N |
1234 | r10_bio->devs[i].bio = bio; |
1235 | atomic_inc(&rdev->nr_pending); | |
475b0321 N |
1236 | if (rrdev) { |
1237 | r10_bio->devs[i].repl_bio = bio; | |
1238 | atomic_inc(&rrdev->nr_pending); | |
1239 | } | |
1da177e4 LT |
1240 | } |
1241 | rcu_read_unlock(); | |
1242 | ||
6bfe0b49 DW |
1243 | if (unlikely(blocked_rdev)) { |
1244 | /* Have to wait for this device to get unblocked, then retry */ | |
1245 | int j; | |
1246 | int d; | |
1247 | ||
475b0321 | 1248 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1249 | if (r10_bio->devs[j].bio) { |
1250 | d = r10_bio->devs[j].devnum; | |
1251 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1252 | } | |
475b0321 | 1253 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1254 | struct md_rdev *rdev; |
475b0321 | 1255 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1256 | rdev = conf->mirrors[d].replacement; |
1257 | if (!rdev) { | |
1258 | /* Race with remove_disk */ | |
1259 | smp_mb(); | |
1260 | rdev = conf->mirrors[d].rdev; | |
1261 | } | |
1262 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1263 | } |
1264 | } | |
6bfe0b49 DW |
1265 | allow_barrier(conf); |
1266 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1267 | wait_barrier(conf); | |
1268 | goto retry_write; | |
1269 | } | |
1270 | ||
d4432c23 N |
1271 | if (max_sectors < r10_bio->sectors) { |
1272 | /* We are splitting this into multiple parts, so | |
1273 | * we need to prepare for allocating another r10_bio. | |
1274 | */ | |
1275 | r10_bio->sectors = max_sectors; | |
1276 | spin_lock_irq(&conf->device_lock); | |
1277 | if (bio->bi_phys_segments == 0) | |
1278 | bio->bi_phys_segments = 2; | |
1279 | else | |
1280 | bio->bi_phys_segments++; | |
1281 | spin_unlock_irq(&conf->device_lock); | |
1282 | } | |
1283 | sectors_handled = r10_bio->sector + max_sectors - bio->bi_sector; | |
1284 | ||
4e78064f | 1285 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1286 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1287 | |
1da177e4 LT |
1288 | for (i = 0; i < conf->copies; i++) { |
1289 | struct bio *mbio; | |
1290 | int d = r10_bio->devs[i].devnum; | |
1291 | if (!r10_bio->devs[i].bio) | |
1292 | continue; | |
1293 | ||
a167f663 | 1294 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
d4432c23 N |
1295 | md_trim_bio(mbio, r10_bio->sector - bio->bi_sector, |
1296 | max_sectors); | |
1da177e4 LT |
1297 | r10_bio->devs[i].bio = mbio; |
1298 | ||
d4432c23 N |
1299 | mbio->bi_sector = (r10_bio->devs[i].addr+ |
1300 | conf->mirrors[d].rdev->data_offset); | |
1da177e4 LT |
1301 | mbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
1302 | mbio->bi_end_io = raid10_end_write_request; | |
e9c7469b | 1303 | mbio->bi_rw = WRITE | do_sync | do_fua; |
1da177e4 LT |
1304 | mbio->bi_private = r10_bio; |
1305 | ||
1306 | atomic_inc(&r10_bio->remaining); | |
4e78064f N |
1307 | spin_lock_irqsave(&conf->device_lock, flags); |
1308 | bio_list_add(&conf->pending_bio_list, mbio); | |
34db0cd6 | 1309 | conf->pending_count++; |
4e78064f | 1310 | spin_unlock_irqrestore(&conf->device_lock, flags); |
475b0321 N |
1311 | |
1312 | if (!r10_bio->devs[i].repl_bio) | |
1313 | continue; | |
1314 | ||
1315 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
1316 | md_trim_bio(mbio, r10_bio->sector - bio->bi_sector, | |
1317 | max_sectors); | |
1318 | r10_bio->devs[i].repl_bio = mbio; | |
1319 | ||
4ca40c2c N |
1320 | /* We are actively writing to the original device |
1321 | * so it cannot disappear, so the replacement cannot | |
1322 | * become NULL here | |
1323 | */ | |
475b0321 N |
1324 | mbio->bi_sector = (r10_bio->devs[i].addr+ |
1325 | conf->mirrors[d].replacement->data_offset); | |
1326 | mbio->bi_bdev = conf->mirrors[d].replacement->bdev; | |
1327 | mbio->bi_end_io = raid10_end_write_request; | |
1328 | mbio->bi_rw = WRITE | do_sync | do_fua; | |
1329 | mbio->bi_private = r10_bio; | |
1330 | ||
1331 | atomic_inc(&r10_bio->remaining); | |
1332 | spin_lock_irqsave(&conf->device_lock, flags); | |
1333 | bio_list_add(&conf->pending_bio_list, mbio); | |
1334 | conf->pending_count++; | |
1335 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1336 | } |
1337 | ||
079fa166 N |
1338 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1339 | * after checking if we need to go around again. | |
1340 | */ | |
a35e63ef | 1341 | |
d4432c23 | 1342 | if (sectors_handled < (bio->bi_size >> 9)) { |
079fa166 | 1343 | one_write_done(r10_bio); |
5e570289 | 1344 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1345 | * in bio->bi_phys_segments. |
1346 | */ | |
1347 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1348 | ||
1349 | r10_bio->master_bio = bio; | |
1350 | r10_bio->sectors = (bio->bi_size >> 9) - sectors_handled; | |
1351 | ||
1352 | r10_bio->mddev = mddev; | |
1353 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1354 | r10_bio->state = 0; | |
1355 | goto retry_write; | |
1356 | } | |
079fa166 N |
1357 | one_write_done(r10_bio); |
1358 | ||
1359 | /* In case raid10d snuck in to freeze_array */ | |
1360 | wake_up(&conf->wait_barrier); | |
d4432c23 | 1361 | |
c3b328ac | 1362 | if (do_sync || !mddev->bitmap || !plugged) |
e3881a68 | 1363 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
1364 | } |
1365 | ||
fd01b88c | 1366 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1367 | { |
e879a879 | 1368 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1369 | int i; |
1370 | ||
1371 | if (conf->near_copies < conf->raid_disks) | |
9d8f0363 | 1372 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
1373 | if (conf->near_copies > 1) |
1374 | seq_printf(seq, " %d near-copies", conf->near_copies); | |
c93983bf N |
1375 | if (conf->far_copies > 1) { |
1376 | if (conf->far_offset) | |
1377 | seq_printf(seq, " %d offset-copies", conf->far_copies); | |
1378 | else | |
1379 | seq_printf(seq, " %d far-copies", conf->far_copies); | |
1380 | } | |
1da177e4 | 1381 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
76186dd8 | 1382 | conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1383 | for (i = 0; i < conf->raid_disks; i++) |
1384 | seq_printf(seq, "%s", | |
1385 | conf->mirrors[i].rdev && | |
b2d444d7 | 1386 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
1387 | seq_printf(seq, "]"); |
1388 | } | |
1389 | ||
700c7213 N |
1390 | /* check if there are enough drives for |
1391 | * every block to appear on atleast one. | |
1392 | * Don't consider the device numbered 'ignore' | |
1393 | * as we might be about to remove it. | |
1394 | */ | |
e879a879 | 1395 | static int enough(struct r10conf *conf, int ignore) |
700c7213 N |
1396 | { |
1397 | int first = 0; | |
1398 | ||
1399 | do { | |
1400 | int n = conf->copies; | |
1401 | int cnt = 0; | |
1402 | while (n--) { | |
1403 | if (conf->mirrors[first].rdev && | |
1404 | first != ignore) | |
1405 | cnt++; | |
1406 | first = (first+1) % conf->raid_disks; | |
1407 | } | |
1408 | if (cnt == 0) | |
1409 | return 0; | |
1410 | } while (first != 0); | |
1411 | return 1; | |
1412 | } | |
1413 | ||
fd01b88c | 1414 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1415 | { |
1416 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1417 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1418 | |
1419 | /* | |
1420 | * If it is not operational, then we have already marked it as dead | |
1421 | * else if it is the last working disks, ignore the error, let the | |
1422 | * next level up know. | |
1423 | * else mark the drive as failed | |
1424 | */ | |
b2d444d7 | 1425 | if (test_bit(In_sync, &rdev->flags) |
700c7213 | 1426 | && !enough(conf, rdev->raid_disk)) |
1da177e4 LT |
1427 | /* |
1428 | * Don't fail the drive, just return an IO error. | |
1da177e4 LT |
1429 | */ |
1430 | return; | |
c04be0aa N |
1431 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1432 | unsigned long flags; | |
1433 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1434 | mddev->degraded++; |
c04be0aa | 1435 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1436 | /* |
1437 | * if recovery is running, make sure it aborts. | |
1438 | */ | |
dfc70645 | 1439 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1440 | } |
de393cde | 1441 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1442 | set_bit(Faulty, &rdev->flags); |
850b2b42 | 1443 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1444 | printk(KERN_ALERT |
1445 | "md/raid10:%s: Disk failure on %s, disabling device.\n" | |
1446 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
128595ed N |
1447 | mdname(mddev), bdevname(rdev->bdev, b), |
1448 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1449 | } |
1450 | ||
e879a879 | 1451 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1452 | { |
1453 | int i; | |
0f6d02d5 | 1454 | struct mirror_info *tmp; |
1da177e4 | 1455 | |
128595ed | 1456 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1457 | if (!conf) { |
128595ed | 1458 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1459 | return; |
1460 | } | |
128595ed | 1461 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1462 | conf->raid_disks); |
1463 | ||
1464 | for (i = 0; i < conf->raid_disks; i++) { | |
1465 | char b[BDEVNAME_SIZE]; | |
1466 | tmp = conf->mirrors + i; | |
1467 | if (tmp->rdev) | |
128595ed | 1468 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1469 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1470 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1471 | bdevname(tmp->rdev->bdev,b)); |
1472 | } | |
1473 | } | |
1474 | ||
e879a879 | 1475 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1476 | { |
0a27ec96 N |
1477 | wait_barrier(conf); |
1478 | allow_barrier(conf); | |
1da177e4 LT |
1479 | |
1480 | mempool_destroy(conf->r10buf_pool); | |
1481 | conf->r10buf_pool = NULL; | |
1482 | } | |
1483 | ||
fd01b88c | 1484 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1485 | { |
1486 | int i; | |
e879a879 | 1487 | struct r10conf *conf = mddev->private; |
0f6d02d5 | 1488 | struct mirror_info *tmp; |
6b965620 N |
1489 | int count = 0; |
1490 | unsigned long flags; | |
1da177e4 LT |
1491 | |
1492 | /* | |
1493 | * Find all non-in_sync disks within the RAID10 configuration | |
1494 | * and mark them in_sync | |
1495 | */ | |
1496 | for (i = 0; i < conf->raid_disks; i++) { | |
1497 | tmp = conf->mirrors + i; | |
4ca40c2c N |
1498 | if (tmp->replacement |
1499 | && tmp->replacement->recovery_offset == MaxSector | |
1500 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1501 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1502 | /* Replacement has just become active */ | |
1503 | if (!tmp->rdev | |
1504 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1505 | count++; | |
1506 | if (tmp->rdev) { | |
1507 | /* Replaced device not technically faulty, | |
1508 | * but we need to be sure it gets removed | |
1509 | * and never re-added. | |
1510 | */ | |
1511 | set_bit(Faulty, &tmp->rdev->flags); | |
1512 | sysfs_notify_dirent_safe( | |
1513 | tmp->rdev->sysfs_state); | |
1514 | } | |
1515 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1516 | } else if (tmp->rdev | |
1517 | && !test_bit(Faulty, &tmp->rdev->flags) | |
1518 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1519 | count++; |
e6ffbcb6 | 1520 | sysfs_notify_dirent(tmp->rdev->sysfs_state); |
1da177e4 LT |
1521 | } |
1522 | } | |
6b965620 N |
1523 | spin_lock_irqsave(&conf->device_lock, flags); |
1524 | mddev->degraded -= count; | |
1525 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1526 | |
1527 | print_conf(conf); | |
6b965620 | 1528 | return count; |
1da177e4 LT |
1529 | } |
1530 | ||
1531 | ||
fd01b88c | 1532 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1533 | { |
e879a879 | 1534 | struct r10conf *conf = mddev->private; |
199050ea | 1535 | int err = -EEXIST; |
1da177e4 | 1536 | int mirror; |
6c2fce2e | 1537 | int first = 0; |
84707f38 | 1538 | int last = conf->raid_disks - 1; |
050b6615 | 1539 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
1540 | |
1541 | if (mddev->recovery_cp < MaxSector) | |
1542 | /* only hot-add to in-sync arrays, as recovery is | |
1543 | * very different from resync | |
1544 | */ | |
199050ea | 1545 | return -EBUSY; |
dc10c643 | 1546 | if (rdev->saved_raid_disk < 0 && !enough(conf, -1)) |
199050ea | 1547 | return -EINVAL; |
1da177e4 | 1548 | |
a53a6c85 | 1549 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1550 | first = last = rdev->raid_disk; |
1da177e4 | 1551 | |
050b6615 N |
1552 | if (q->merge_bvec_fn) { |
1553 | set_bit(Unmerged, &rdev->flags); | |
1554 | mddev->merge_check_needed = 1; | |
1555 | } | |
1556 | ||
2c4193df | 1557 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1558 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1559 | mirror = rdev->saved_raid_disk; | |
1560 | else | |
6c2fce2e | 1561 | mirror = first; |
2bb77736 | 1562 | for ( ; mirror <= last ; mirror++) { |
0f6d02d5 | 1563 | struct mirror_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1564 | if (p->recovery_disabled == mddev->recovery_disabled) |
1565 | continue; | |
b7044d41 N |
1566 | if (p->rdev) { |
1567 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1568 | p->replacement != NULL) | |
1569 | continue; | |
1570 | clear_bit(In_sync, &rdev->flags); | |
1571 | set_bit(Replacement, &rdev->flags); | |
1572 | rdev->raid_disk = mirror; | |
1573 | err = 0; | |
1574 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1575 | rdev->data_offset << 9); | |
b7044d41 N |
1576 | conf->fullsync = 1; |
1577 | rcu_assign_pointer(p->replacement, rdev); | |
1578 | break; | |
1579 | } | |
1da177e4 | 1580 | |
2bb77736 N |
1581 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1582 | rdev->data_offset << 9); | |
1da177e4 | 1583 | |
2bb77736 | 1584 | p->head_position = 0; |
d890fa2b | 1585 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1586 | rdev->raid_disk = mirror; |
1587 | err = 0; | |
1588 | if (rdev->saved_raid_disk != mirror) | |
1589 | conf->fullsync = 1; | |
1590 | rcu_assign_pointer(p->rdev, rdev); | |
1591 | break; | |
1592 | } | |
050b6615 N |
1593 | if (err == 0 && test_bit(Unmerged, &rdev->flags)) { |
1594 | /* Some requests might not have seen this new | |
1595 | * merge_bvec_fn. We must wait for them to complete | |
1596 | * before merging the device fully. | |
1597 | * First we make sure any code which has tested | |
1598 | * our function has submitted the request, then | |
1599 | * we wait for all outstanding requests to complete. | |
1600 | */ | |
1601 | synchronize_sched(); | |
1602 | raise_barrier(conf, 0); | |
1603 | lower_barrier(conf); | |
1604 | clear_bit(Unmerged, &rdev->flags); | |
1605 | } | |
ac5e7113 | 1606 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1607 | print_conf(conf); |
199050ea | 1608 | return err; |
1da177e4 LT |
1609 | } |
1610 | ||
b8321b68 | 1611 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1612 | { |
e879a879 | 1613 | struct r10conf *conf = mddev->private; |
1da177e4 | 1614 | int err = 0; |
b8321b68 | 1615 | int number = rdev->raid_disk; |
c8ab903e N |
1616 | struct md_rdev **rdevp; |
1617 | struct mirror_info *p = conf->mirrors + number; | |
1da177e4 LT |
1618 | |
1619 | print_conf(conf); | |
c8ab903e N |
1620 | if (rdev == p->rdev) |
1621 | rdevp = &p->rdev; | |
1622 | else if (rdev == p->replacement) | |
1623 | rdevp = &p->replacement; | |
1624 | else | |
1625 | return 0; | |
1626 | ||
1627 | if (test_bit(In_sync, &rdev->flags) || | |
1628 | atomic_read(&rdev->nr_pending)) { | |
1629 | err = -EBUSY; | |
1630 | goto abort; | |
1631 | } | |
1632 | /* Only remove faulty devices if recovery | |
1633 | * is not possible. | |
1634 | */ | |
1635 | if (!test_bit(Faulty, &rdev->flags) && | |
1636 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1637 | (!p->replacement || p->replacement == rdev) && |
c8ab903e N |
1638 | enough(conf, -1)) { |
1639 | err = -EBUSY; | |
1640 | goto abort; | |
1da177e4 | 1641 | } |
c8ab903e N |
1642 | *rdevp = NULL; |
1643 | synchronize_rcu(); | |
1644 | if (atomic_read(&rdev->nr_pending)) { | |
1645 | /* lost the race, try later */ | |
1646 | err = -EBUSY; | |
1647 | *rdevp = rdev; | |
1648 | goto abort; | |
4ca40c2c N |
1649 | } else if (p->replacement) { |
1650 | /* We must have just cleared 'rdev' */ | |
1651 | p->rdev = p->replacement; | |
1652 | clear_bit(Replacement, &p->replacement->flags); | |
1653 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1654 | * but will never see neither -- if they are careful. | |
1655 | */ | |
1656 | p->replacement = NULL; | |
1657 | clear_bit(WantReplacement, &rdev->flags); | |
1658 | } else | |
1659 | /* We might have just remove the Replacement as faulty | |
1660 | * Clear the flag just in case | |
1661 | */ | |
1662 | clear_bit(WantReplacement, &rdev->flags); | |
1663 | ||
c8ab903e N |
1664 | err = md_integrity_register(mddev); |
1665 | ||
1da177e4 LT |
1666 | abort: |
1667 | ||
1668 | print_conf(conf); | |
1669 | return err; | |
1670 | } | |
1671 | ||
1672 | ||
6712ecf8 | 1673 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1674 | { |
9f2c9d12 | 1675 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1676 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1677 | int d; |
1da177e4 | 1678 | |
69335ef3 | 1679 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); |
0eb3ff12 N |
1680 | |
1681 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1682 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
e684e41d N |
1683 | else |
1684 | /* The write handler will notice the lack of | |
1685 | * R10BIO_Uptodate and record any errors etc | |
1686 | */ | |
4dbcdc75 N |
1687 | atomic_add(r10_bio->sectors, |
1688 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1689 | |
1690 | /* for reconstruct, we always reschedule after a read. | |
1691 | * for resync, only after all reads | |
1692 | */ | |
73d5c38a | 1693 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1694 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1695 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1696 | /* we have read all the blocks, | |
1697 | * do the comparison in process context in raid10d | |
1698 | */ | |
1699 | reschedule_retry(r10_bio); | |
1700 | } | |
1da177e4 LT |
1701 | } |
1702 | ||
9f2c9d12 | 1703 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1704 | { |
fd01b88c | 1705 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1706 | |
1da177e4 LT |
1707 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1708 | if (r10_bio->master_bio == NULL) { | |
1709 | /* the primary of several recovery bios */ | |
73d5c38a | 1710 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1711 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1712 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1713 | reschedule_retry(r10_bio); |
1714 | else | |
1715 | put_buf(r10_bio); | |
73d5c38a | 1716 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1717 | break; |
1718 | } else { | |
9f2c9d12 | 1719 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1720 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1721 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1722 | reschedule_retry(r10_bio); |
1723 | else | |
1724 | put_buf(r10_bio); | |
1da177e4 LT |
1725 | r10_bio = r10_bio2; |
1726 | } | |
1727 | } | |
1da177e4 LT |
1728 | } |
1729 | ||
5e570289 N |
1730 | static void end_sync_write(struct bio *bio, int error) |
1731 | { | |
1732 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1733 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 1734 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1735 | struct r10conf *conf = mddev->private; |
5e570289 N |
1736 | int d; |
1737 | sector_t first_bad; | |
1738 | int bad_sectors; | |
1739 | int slot; | |
9ad1aefc | 1740 | int repl; |
4ca40c2c | 1741 | struct md_rdev *rdev = NULL; |
5e570289 | 1742 | |
9ad1aefc N |
1743 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1744 | if (repl) | |
1745 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1746 | else |
9ad1aefc | 1747 | rdev = conf->mirrors[d].rdev; |
5e570289 N |
1748 | |
1749 | if (!uptodate) { | |
9ad1aefc N |
1750 | if (repl) |
1751 | md_error(mddev, rdev); | |
1752 | else { | |
1753 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1754 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1755 | set_bit(MD_RECOVERY_NEEDED, | |
1756 | &rdev->mddev->recovery); | |
9ad1aefc N |
1757 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1758 | } | |
1759 | } else if (is_badblock(rdev, | |
5e570289 N |
1760 | r10_bio->devs[slot].addr, |
1761 | r10_bio->sectors, | |
1762 | &first_bad, &bad_sectors)) | |
1763 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1764 | ||
9ad1aefc | 1765 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1766 | |
1767 | end_sync_request(r10_bio); | |
1768 | } | |
1769 | ||
1da177e4 LT |
1770 | /* |
1771 | * Note: sync and recover and handled very differently for raid10 | |
1772 | * This code is for resync. | |
1773 | * For resync, we read through virtual addresses and read all blocks. | |
1774 | * If there is any error, we schedule a write. The lowest numbered | |
1775 | * drive is authoritative. | |
1776 | * However requests come for physical address, so we need to map. | |
1777 | * For every physical address there are raid_disks/copies virtual addresses, | |
1778 | * which is always are least one, but is not necessarly an integer. | |
1779 | * This means that a physical address can span multiple chunks, so we may | |
1780 | * have to submit multiple io requests for a single sync request. | |
1781 | */ | |
1782 | /* | |
1783 | * We check if all blocks are in-sync and only write to blocks that | |
1784 | * aren't in sync | |
1785 | */ | |
9f2c9d12 | 1786 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1787 | { |
e879a879 | 1788 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1789 | int i, first; |
1790 | struct bio *tbio, *fbio; | |
1791 | ||
1792 | atomic_set(&r10_bio->remaining, 1); | |
1793 | ||
1794 | /* find the first device with a block */ | |
1795 | for (i=0; i<conf->copies; i++) | |
1796 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) | |
1797 | break; | |
1798 | ||
1799 | if (i == conf->copies) | |
1800 | goto done; | |
1801 | ||
1802 | first = i; | |
1803 | fbio = r10_bio->devs[i].bio; | |
1804 | ||
1805 | /* now find blocks with errors */ | |
0eb3ff12 N |
1806 | for (i=0 ; i < conf->copies ; i++) { |
1807 | int j, d; | |
1808 | int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); | |
1da177e4 | 1809 | |
1da177e4 | 1810 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1811 | |
1812 | if (tbio->bi_end_io != end_sync_read) | |
1813 | continue; | |
1814 | if (i == first) | |
1da177e4 | 1815 | continue; |
0eb3ff12 N |
1816 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) { |
1817 | /* We know that the bi_io_vec layout is the same for | |
1818 | * both 'first' and 'i', so we just compare them. | |
1819 | * All vec entries are PAGE_SIZE; | |
1820 | */ | |
1821 | for (j = 0; j < vcnt; j++) | |
1822 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), | |
1823 | page_address(tbio->bi_io_vec[j].bv_page), | |
1824 | PAGE_SIZE)) | |
1825 | break; | |
1826 | if (j == vcnt) | |
1827 | continue; | |
1828 | mddev->resync_mismatches += r10_bio->sectors; | |
f84ee364 N |
1829 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1830 | /* Don't fix anything. */ | |
1831 | continue; | |
0eb3ff12 | 1832 | } |
f84ee364 N |
1833 | /* Ok, we need to write this bio, either to correct an |
1834 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
1835 | * First we need to fixup bv_offset, bv_len and |
1836 | * bi_vecs, as the read request might have corrupted these | |
1837 | */ | |
1838 | tbio->bi_vcnt = vcnt; | |
1839 | tbio->bi_size = r10_bio->sectors << 9; | |
1840 | tbio->bi_idx = 0; | |
1841 | tbio->bi_phys_segments = 0; | |
1da177e4 LT |
1842 | tbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1843 | tbio->bi_flags |= 1 << BIO_UPTODATE; | |
1844 | tbio->bi_next = NULL; | |
1845 | tbio->bi_rw = WRITE; | |
1846 | tbio->bi_private = r10_bio; | |
1847 | tbio->bi_sector = r10_bio->devs[i].addr; | |
1848 | ||
1849 | for (j=0; j < vcnt ; j++) { | |
1850 | tbio->bi_io_vec[j].bv_offset = 0; | |
1851 | tbio->bi_io_vec[j].bv_len = PAGE_SIZE; | |
1852 | ||
1853 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1854 | page_address(fbio->bi_io_vec[j].bv_page), | |
1855 | PAGE_SIZE); | |
1856 | } | |
1857 | tbio->bi_end_io = end_sync_write; | |
1858 | ||
1859 | d = r10_bio->devs[i].devnum; | |
1860 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1861 | atomic_inc(&r10_bio->remaining); | |
1862 | md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); | |
1863 | ||
1864 | tbio->bi_sector += conf->mirrors[d].rdev->data_offset; | |
1865 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1866 | generic_make_request(tbio); | |
1867 | } | |
1868 | ||
9ad1aefc N |
1869 | /* Now write out to any replacement devices |
1870 | * that are active | |
1871 | */ | |
1872 | for (i = 0; i < conf->copies; i++) { | |
1873 | int j, d; | |
1874 | int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); | |
1875 | ||
1876 | tbio = r10_bio->devs[i].repl_bio; | |
1877 | if (!tbio || !tbio->bi_end_io) | |
1878 | continue; | |
1879 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
1880 | && r10_bio->devs[i].bio != fbio) | |
1881 | for (j = 0; j < vcnt; j++) | |
1882 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1883 | page_address(fbio->bi_io_vec[j].bv_page), | |
1884 | PAGE_SIZE); | |
1885 | d = r10_bio->devs[i].devnum; | |
1886 | atomic_inc(&r10_bio->remaining); | |
1887 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
1888 | tbio->bi_size >> 9); | |
1889 | generic_make_request(tbio); | |
1890 | } | |
1891 | ||
1da177e4 LT |
1892 | done: |
1893 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
1894 | md_done_sync(mddev, r10_bio->sectors, 1); | |
1895 | put_buf(r10_bio); | |
1896 | } | |
1897 | } | |
1898 | ||
1899 | /* | |
1900 | * Now for the recovery code. | |
1901 | * Recovery happens across physical sectors. | |
1902 | * We recover all non-is_sync drives by finding the virtual address of | |
1903 | * each, and then choose a working drive that also has that virt address. | |
1904 | * There is a separate r10_bio for each non-in_sync drive. | |
1905 | * Only the first two slots are in use. The first for reading, | |
1906 | * The second for writing. | |
1907 | * | |
1908 | */ | |
9f2c9d12 | 1909 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
1910 | { |
1911 | /* We got a read error during recovery. | |
1912 | * We repeat the read in smaller page-sized sections. | |
1913 | * If a read succeeds, write it to the new device or record | |
1914 | * a bad block if we cannot. | |
1915 | * If a read fails, record a bad block on both old and | |
1916 | * new devices. | |
1917 | */ | |
fd01b88c | 1918 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1919 | struct r10conf *conf = mddev->private; |
5e570289 N |
1920 | struct bio *bio = r10_bio->devs[0].bio; |
1921 | sector_t sect = 0; | |
1922 | int sectors = r10_bio->sectors; | |
1923 | int idx = 0; | |
1924 | int dr = r10_bio->devs[0].devnum; | |
1925 | int dw = r10_bio->devs[1].devnum; | |
1926 | ||
1927 | while (sectors) { | |
1928 | int s = sectors; | |
3cb03002 | 1929 | struct md_rdev *rdev; |
5e570289 N |
1930 | sector_t addr; |
1931 | int ok; | |
1932 | ||
1933 | if (s > (PAGE_SIZE>>9)) | |
1934 | s = PAGE_SIZE >> 9; | |
1935 | ||
1936 | rdev = conf->mirrors[dr].rdev; | |
1937 | addr = r10_bio->devs[0].addr + sect, | |
1938 | ok = sync_page_io(rdev, | |
1939 | addr, | |
1940 | s << 9, | |
1941 | bio->bi_io_vec[idx].bv_page, | |
1942 | READ, false); | |
1943 | if (ok) { | |
1944 | rdev = conf->mirrors[dw].rdev; | |
1945 | addr = r10_bio->devs[1].addr + sect; | |
1946 | ok = sync_page_io(rdev, | |
1947 | addr, | |
1948 | s << 9, | |
1949 | bio->bi_io_vec[idx].bv_page, | |
1950 | WRITE, false); | |
b7044d41 | 1951 | if (!ok) { |
5e570289 | 1952 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
1953 | if (!test_and_set_bit(WantReplacement, |
1954 | &rdev->flags)) | |
1955 | set_bit(MD_RECOVERY_NEEDED, | |
1956 | &rdev->mddev->recovery); | |
1957 | } | |
5e570289 N |
1958 | } |
1959 | if (!ok) { | |
1960 | /* We don't worry if we cannot set a bad block - | |
1961 | * it really is bad so there is no loss in not | |
1962 | * recording it yet | |
1963 | */ | |
1964 | rdev_set_badblocks(rdev, addr, s, 0); | |
1965 | ||
1966 | if (rdev != conf->mirrors[dw].rdev) { | |
1967 | /* need bad block on destination too */ | |
3cb03002 | 1968 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
1969 | addr = r10_bio->devs[1].addr + sect; |
1970 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
1971 | if (!ok) { | |
1972 | /* just abort the recovery */ | |
1973 | printk(KERN_NOTICE | |
1974 | "md/raid10:%s: recovery aborted" | |
1975 | " due to read error\n", | |
1976 | mdname(mddev)); | |
1977 | ||
1978 | conf->mirrors[dw].recovery_disabled | |
1979 | = mddev->recovery_disabled; | |
1980 | set_bit(MD_RECOVERY_INTR, | |
1981 | &mddev->recovery); | |
1982 | break; | |
1983 | } | |
1984 | } | |
1985 | } | |
1986 | ||
1987 | sectors -= s; | |
1988 | sect += s; | |
1989 | idx++; | |
1990 | } | |
1991 | } | |
1da177e4 | 1992 | |
9f2c9d12 | 1993 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1994 | { |
e879a879 | 1995 | struct r10conf *conf = mddev->private; |
c65060ad | 1996 | int d; |
24afd80d | 1997 | struct bio *wbio, *wbio2; |
1da177e4 | 1998 | |
5e570289 N |
1999 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2000 | fix_recovery_read_error(r10_bio); | |
2001 | end_sync_request(r10_bio); | |
2002 | return; | |
2003 | } | |
2004 | ||
c65060ad NK |
2005 | /* |
2006 | * share the pages with the first bio | |
1da177e4 LT |
2007 | * and submit the write request |
2008 | */ | |
1da177e4 | 2009 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2010 | wbio = r10_bio->devs[1].bio; |
2011 | wbio2 = r10_bio->devs[1].repl_bio; | |
2012 | if (wbio->bi_end_io) { | |
2013 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2014 | md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); | |
2015 | generic_make_request(wbio); | |
2016 | } | |
2017 | if (wbio2 && wbio2->bi_end_io) { | |
2018 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); | |
2019 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
2020 | wbio2->bi_size >> 9); | |
2021 | generic_make_request(wbio2); | |
2022 | } | |
1da177e4 LT |
2023 | } |
2024 | ||
2025 | ||
1e50915f RB |
2026 | /* |
2027 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2028 | * We halve the read error count for every hour that has elapsed | |
2029 | * since the last recorded read error. | |
2030 | * | |
2031 | */ | |
fd01b88c | 2032 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f RB |
2033 | { |
2034 | struct timespec cur_time_mon; | |
2035 | unsigned long hours_since_last; | |
2036 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2037 | ||
2038 | ktime_get_ts(&cur_time_mon); | |
2039 | ||
2040 | if (rdev->last_read_error.tv_sec == 0 && | |
2041 | rdev->last_read_error.tv_nsec == 0) { | |
2042 | /* first time we've seen a read error */ | |
2043 | rdev->last_read_error = cur_time_mon; | |
2044 | return; | |
2045 | } | |
2046 | ||
2047 | hours_since_last = (cur_time_mon.tv_sec - | |
2048 | rdev->last_read_error.tv_sec) / 3600; | |
2049 | ||
2050 | rdev->last_read_error = cur_time_mon; | |
2051 | ||
2052 | /* | |
2053 | * if hours_since_last is > the number of bits in read_errors | |
2054 | * just set read errors to 0. We do this to avoid | |
2055 | * overflowing the shift of read_errors by hours_since_last. | |
2056 | */ | |
2057 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2058 | atomic_set(&rdev->read_errors, 0); | |
2059 | else | |
2060 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2061 | } | |
2062 | ||
3cb03002 | 2063 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2064 | int sectors, struct page *page, int rw) |
2065 | { | |
2066 | sector_t first_bad; | |
2067 | int bad_sectors; | |
2068 | ||
2069 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2070 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2071 | return -1; | |
2072 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
2073 | /* success */ | |
2074 | return 1; | |
b7044d41 | 2075 | if (rw == WRITE) { |
58c54fcc | 2076 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2077 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2078 | set_bit(MD_RECOVERY_NEEDED, | |
2079 | &rdev->mddev->recovery); | |
2080 | } | |
58c54fcc N |
2081 | /* need to record an error - either for the block or the device */ |
2082 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2083 | md_error(rdev->mddev, rdev); | |
2084 | return 0; | |
2085 | } | |
2086 | ||
1da177e4 LT |
2087 | /* |
2088 | * This is a kernel thread which: | |
2089 | * | |
2090 | * 1. Retries failed read operations on working mirrors. | |
2091 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2092 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2093 | */ |
2094 | ||
e879a879 | 2095 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2096 | { |
2097 | int sect = 0; /* Offset from r10_bio->sector */ | |
2098 | int sectors = r10_bio->sectors; | |
3cb03002 | 2099 | struct md_rdev*rdev; |
1e50915f | 2100 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2101 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2102 | |
7c4e06ff N |
2103 | /* still own a reference to this rdev, so it cannot |
2104 | * have been cleared recently. | |
2105 | */ | |
2106 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2107 | |
7c4e06ff N |
2108 | if (test_bit(Faulty, &rdev->flags)) |
2109 | /* drive has already been failed, just ignore any | |
2110 | more fix_read_error() attempts */ | |
2111 | return; | |
1e50915f | 2112 | |
7c4e06ff N |
2113 | check_decay_read_errors(mddev, rdev); |
2114 | atomic_inc(&rdev->read_errors); | |
2115 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2116 | char b[BDEVNAME_SIZE]; | |
2117 | bdevname(rdev->bdev, b); | |
1e50915f | 2118 | |
7c4e06ff N |
2119 | printk(KERN_NOTICE |
2120 | "md/raid10:%s: %s: Raid device exceeded " | |
2121 | "read_error threshold [cur %d:max %d]\n", | |
2122 | mdname(mddev), b, | |
2123 | atomic_read(&rdev->read_errors), max_read_errors); | |
2124 | printk(KERN_NOTICE | |
2125 | "md/raid10:%s: %s: Failing raid device\n", | |
2126 | mdname(mddev), b); | |
2127 | md_error(mddev, conf->mirrors[d].rdev); | |
fae8cc5e | 2128 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2129 | return; |
1e50915f | 2130 | } |
1e50915f | 2131 | |
6814d536 N |
2132 | while(sectors) { |
2133 | int s = sectors; | |
2134 | int sl = r10_bio->read_slot; | |
2135 | int success = 0; | |
2136 | int start; | |
2137 | ||
2138 | if (s > (PAGE_SIZE>>9)) | |
2139 | s = PAGE_SIZE >> 9; | |
2140 | ||
2141 | rcu_read_lock(); | |
2142 | do { | |
8dbed5ce N |
2143 | sector_t first_bad; |
2144 | int bad_sectors; | |
2145 | ||
0544a21d | 2146 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2147 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2148 | if (rdev && | |
050b6615 | 2149 | !test_bit(Unmerged, &rdev->flags) && |
8dbed5ce N |
2150 | test_bit(In_sync, &rdev->flags) && |
2151 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, | |
2152 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2153 | atomic_inc(&rdev->nr_pending); |
2154 | rcu_read_unlock(); | |
2b193363 | 2155 | success = sync_page_io(rdev, |
6814d536 | 2156 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2157 | sect, |
6814d536 | 2158 | s<<9, |
ccebd4c4 | 2159 | conf->tmppage, READ, false); |
6814d536 N |
2160 | rdev_dec_pending(rdev, mddev); |
2161 | rcu_read_lock(); | |
2162 | if (success) | |
2163 | break; | |
2164 | } | |
2165 | sl++; | |
2166 | if (sl == conf->copies) | |
2167 | sl = 0; | |
2168 | } while (!success && sl != r10_bio->read_slot); | |
2169 | rcu_read_unlock(); | |
2170 | ||
2171 | if (!success) { | |
58c54fcc N |
2172 | /* Cannot read from anywhere, just mark the block |
2173 | * as bad on the first device to discourage future | |
2174 | * reads. | |
2175 | */ | |
6814d536 | 2176 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2177 | rdev = conf->mirrors[dn].rdev; |
2178 | ||
2179 | if (!rdev_set_badblocks( | |
2180 | rdev, | |
2181 | r10_bio->devs[r10_bio->read_slot].addr | |
2182 | + sect, | |
fae8cc5e | 2183 | s, 0)) { |
58c54fcc | 2184 | md_error(mddev, rdev); |
fae8cc5e N |
2185 | r10_bio->devs[r10_bio->read_slot].bio |
2186 | = IO_BLOCKED; | |
2187 | } | |
6814d536 N |
2188 | break; |
2189 | } | |
2190 | ||
2191 | start = sl; | |
2192 | /* write it back and re-read */ | |
2193 | rcu_read_lock(); | |
2194 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2195 | char b[BDEVNAME_SIZE]; |
0544a21d | 2196 | |
6814d536 N |
2197 | if (sl==0) |
2198 | sl = conf->copies; | |
2199 | sl--; | |
2200 | d = r10_bio->devs[sl].devnum; | |
2201 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2202 | if (!rdev || |
050b6615 | 2203 | test_bit(Unmerged, &rdev->flags) || |
1294b9c9 N |
2204 | !test_bit(In_sync, &rdev->flags)) |
2205 | continue; | |
2206 | ||
2207 | atomic_inc(&rdev->nr_pending); | |
2208 | rcu_read_unlock(); | |
58c54fcc N |
2209 | if (r10_sync_page_io(rdev, |
2210 | r10_bio->devs[sl].addr + | |
2211 | sect, | |
2212 | s<<9, conf->tmppage, WRITE) | |
1294b9c9 N |
2213 | == 0) { |
2214 | /* Well, this device is dead */ | |
2215 | printk(KERN_NOTICE | |
2216 | "md/raid10:%s: read correction " | |
2217 | "write failed" | |
2218 | " (%d sectors at %llu on %s)\n", | |
2219 | mdname(mddev), s, | |
2220 | (unsigned long long)( | |
2221 | sect + rdev->data_offset), | |
2222 | bdevname(rdev->bdev, b)); | |
2223 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2224 | "drive\n", | |
2225 | mdname(mddev), | |
2226 | bdevname(rdev->bdev, b)); | |
6814d536 | 2227 | } |
1294b9c9 N |
2228 | rdev_dec_pending(rdev, mddev); |
2229 | rcu_read_lock(); | |
6814d536 N |
2230 | } |
2231 | sl = start; | |
2232 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2233 | char b[BDEVNAME_SIZE]; |
0544a21d | 2234 | |
6814d536 N |
2235 | if (sl==0) |
2236 | sl = conf->copies; | |
2237 | sl--; | |
2238 | d = r10_bio->devs[sl].devnum; | |
2239 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
2240 | if (!rdev || |
2241 | !test_bit(In_sync, &rdev->flags)) | |
2242 | continue; | |
6814d536 | 2243 | |
1294b9c9 N |
2244 | atomic_inc(&rdev->nr_pending); |
2245 | rcu_read_unlock(); | |
58c54fcc N |
2246 | switch (r10_sync_page_io(rdev, |
2247 | r10_bio->devs[sl].addr + | |
2248 | sect, | |
2249 | s<<9, conf->tmppage, | |
2250 | READ)) { | |
2251 | case 0: | |
1294b9c9 N |
2252 | /* Well, this device is dead */ |
2253 | printk(KERN_NOTICE | |
2254 | "md/raid10:%s: unable to read back " | |
2255 | "corrected sectors" | |
2256 | " (%d sectors at %llu on %s)\n", | |
2257 | mdname(mddev), s, | |
2258 | (unsigned long long)( | |
2259 | sect + rdev->data_offset), | |
2260 | bdevname(rdev->bdev, b)); | |
2261 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2262 | "drive\n", | |
2263 | mdname(mddev), | |
2264 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2265 | break; |
2266 | case 1: | |
1294b9c9 N |
2267 | printk(KERN_INFO |
2268 | "md/raid10:%s: read error corrected" | |
2269 | " (%d sectors at %llu on %s)\n", | |
2270 | mdname(mddev), s, | |
2271 | (unsigned long long)( | |
2272 | sect + rdev->data_offset), | |
2273 | bdevname(rdev->bdev, b)); | |
2274 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2275 | } |
1294b9c9 N |
2276 | |
2277 | rdev_dec_pending(rdev, mddev); | |
2278 | rcu_read_lock(); | |
6814d536 N |
2279 | } |
2280 | rcu_read_unlock(); | |
2281 | ||
2282 | sectors -= s; | |
2283 | sect += s; | |
2284 | } | |
2285 | } | |
2286 | ||
bd870a16 N |
2287 | static void bi_complete(struct bio *bio, int error) |
2288 | { | |
2289 | complete((struct completion *)bio->bi_private); | |
2290 | } | |
2291 | ||
2292 | static int submit_bio_wait(int rw, struct bio *bio) | |
2293 | { | |
2294 | struct completion event; | |
2295 | rw |= REQ_SYNC; | |
2296 | ||
2297 | init_completion(&event); | |
2298 | bio->bi_private = &event; | |
2299 | bio->bi_end_io = bi_complete; | |
2300 | submit_bio(rw, bio); | |
2301 | wait_for_completion(&event); | |
2302 | ||
2303 | return test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2304 | } | |
2305 | ||
9f2c9d12 | 2306 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2307 | { |
2308 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2309 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2310 | struct r10conf *conf = mddev->private; |
3cb03002 | 2311 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2312 | /* bio has the data to be written to slot 'i' where |
2313 | * we just recently had a write error. | |
2314 | * We repeatedly clone the bio and trim down to one block, | |
2315 | * then try the write. Where the write fails we record | |
2316 | * a bad block. | |
2317 | * It is conceivable that the bio doesn't exactly align with | |
2318 | * blocks. We must handle this. | |
2319 | * | |
2320 | * We currently own a reference to the rdev. | |
2321 | */ | |
2322 | ||
2323 | int block_sectors; | |
2324 | sector_t sector; | |
2325 | int sectors; | |
2326 | int sect_to_write = r10_bio->sectors; | |
2327 | int ok = 1; | |
2328 | ||
2329 | if (rdev->badblocks.shift < 0) | |
2330 | return 0; | |
2331 | ||
2332 | block_sectors = 1 << rdev->badblocks.shift; | |
2333 | sector = r10_bio->sector; | |
2334 | sectors = ((r10_bio->sector + block_sectors) | |
2335 | & ~(sector_t)(block_sectors - 1)) | |
2336 | - sector; | |
2337 | ||
2338 | while (sect_to_write) { | |
2339 | struct bio *wbio; | |
2340 | if (sectors > sect_to_write) | |
2341 | sectors = sect_to_write; | |
2342 | /* Write at 'sector' for 'sectors' */ | |
2343 | wbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
2344 | md_trim_bio(wbio, sector - bio->bi_sector, sectors); | |
2345 | wbio->bi_sector = (r10_bio->devs[i].addr+ | |
2346 | rdev->data_offset+ | |
2347 | (sector - r10_bio->sector)); | |
2348 | wbio->bi_bdev = rdev->bdev; | |
2349 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2350 | /* Failure! */ | |
2351 | ok = rdev_set_badblocks(rdev, sector, | |
2352 | sectors, 0) | |
2353 | && ok; | |
2354 | ||
2355 | bio_put(wbio); | |
2356 | sect_to_write -= sectors; | |
2357 | sector += sectors; | |
2358 | sectors = block_sectors; | |
2359 | } | |
2360 | return ok; | |
2361 | } | |
2362 | ||
9f2c9d12 | 2363 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2364 | { |
2365 | int slot = r10_bio->read_slot; | |
560f8e55 | 2366 | struct bio *bio; |
e879a879 | 2367 | struct r10conf *conf = mddev->private; |
abbf098e | 2368 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2369 | char b[BDEVNAME_SIZE]; |
2370 | unsigned long do_sync; | |
856e08e2 | 2371 | int max_sectors; |
560f8e55 N |
2372 | |
2373 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2374 | * the block and we can fix it. | |
2375 | * We freeze all other IO, and try reading the block from | |
2376 | * other devices. When we find one, we re-write | |
2377 | * and check it that fixes the read error. | |
2378 | * This is all done synchronously while the array is | |
2379 | * frozen. | |
2380 | */ | |
fae8cc5e N |
2381 | bio = r10_bio->devs[slot].bio; |
2382 | bdevname(bio->bi_bdev, b); | |
2383 | bio_put(bio); | |
2384 | r10_bio->devs[slot].bio = NULL; | |
2385 | ||
560f8e55 N |
2386 | if (mddev->ro == 0) { |
2387 | freeze_array(conf); | |
2388 | fix_read_error(conf, mddev, r10_bio); | |
2389 | unfreeze_array(conf); | |
fae8cc5e N |
2390 | } else |
2391 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2392 | ||
abbf098e | 2393 | rdev_dec_pending(rdev, mddev); |
560f8e55 | 2394 | |
7399c31b | 2395 | read_more: |
96c3fd1f N |
2396 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2397 | if (rdev == NULL) { | |
560f8e55 N |
2398 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
2399 | " read error for block %llu\n", | |
7399c31b | 2400 | mdname(mddev), b, |
560f8e55 N |
2401 | (unsigned long long)r10_bio->sector); |
2402 | raid_end_bio_io(r10_bio); | |
560f8e55 N |
2403 | return; |
2404 | } | |
2405 | ||
2406 | do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); | |
560f8e55 | 2407 | slot = r10_bio->read_slot; |
560f8e55 N |
2408 | printk_ratelimited( |
2409 | KERN_ERR | |
2410 | "md/raid10:%s: %s: redirecting" | |
2411 | "sector %llu to another mirror\n", | |
2412 | mdname(mddev), | |
2413 | bdevname(rdev->bdev, b), | |
2414 | (unsigned long long)r10_bio->sector); | |
2415 | bio = bio_clone_mddev(r10_bio->master_bio, | |
2416 | GFP_NOIO, mddev); | |
7399c31b N |
2417 | md_trim_bio(bio, |
2418 | r10_bio->sector - bio->bi_sector, | |
2419 | max_sectors); | |
560f8e55 | 2420 | r10_bio->devs[slot].bio = bio; |
abbf098e | 2421 | r10_bio->devs[slot].rdev = rdev; |
560f8e55 N |
2422 | bio->bi_sector = r10_bio->devs[slot].addr |
2423 | + rdev->data_offset; | |
2424 | bio->bi_bdev = rdev->bdev; | |
2425 | bio->bi_rw = READ | do_sync; | |
2426 | bio->bi_private = r10_bio; | |
2427 | bio->bi_end_io = raid10_end_read_request; | |
7399c31b N |
2428 | if (max_sectors < r10_bio->sectors) { |
2429 | /* Drat - have to split this up more */ | |
2430 | struct bio *mbio = r10_bio->master_bio; | |
2431 | int sectors_handled = | |
2432 | r10_bio->sector + max_sectors | |
2433 | - mbio->bi_sector; | |
2434 | r10_bio->sectors = max_sectors; | |
2435 | spin_lock_irq(&conf->device_lock); | |
2436 | if (mbio->bi_phys_segments == 0) | |
2437 | mbio->bi_phys_segments = 2; | |
2438 | else | |
2439 | mbio->bi_phys_segments++; | |
2440 | spin_unlock_irq(&conf->device_lock); | |
2441 | generic_make_request(bio); | |
7399c31b N |
2442 | |
2443 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2444 | GFP_NOIO); | |
2445 | r10_bio->master_bio = mbio; | |
2446 | r10_bio->sectors = (mbio->bi_size >> 9) | |
2447 | - sectors_handled; | |
2448 | r10_bio->state = 0; | |
2449 | set_bit(R10BIO_ReadError, | |
2450 | &r10_bio->state); | |
2451 | r10_bio->mddev = mddev; | |
2452 | r10_bio->sector = mbio->bi_sector | |
2453 | + sectors_handled; | |
2454 | ||
2455 | goto read_more; | |
2456 | } else | |
2457 | generic_make_request(bio); | |
560f8e55 N |
2458 | } |
2459 | ||
e879a879 | 2460 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2461 | { |
2462 | /* Some sort of write request has finished and it | |
2463 | * succeeded in writing where we thought there was a | |
2464 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2465 | * Or possibly if failed and we need to record |
2466 | * a bad block. | |
749c55e9 N |
2467 | */ |
2468 | int m; | |
3cb03002 | 2469 | struct md_rdev *rdev; |
749c55e9 N |
2470 | |
2471 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2472 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2473 | for (m = 0; m < conf->copies; m++) { |
2474 | int dev = r10_bio->devs[m].devnum; | |
2475 | rdev = conf->mirrors[dev].rdev; | |
2476 | if (r10_bio->devs[m].bio == NULL) | |
2477 | continue; | |
2478 | if (test_bit(BIO_UPTODATE, | |
749c55e9 | 2479 | &r10_bio->devs[m].bio->bi_flags)) { |
749c55e9 N |
2480 | rdev_clear_badblocks( |
2481 | rdev, | |
2482 | r10_bio->devs[m].addr, | |
2483 | r10_bio->sectors); | |
1a0b7cd8 N |
2484 | } else { |
2485 | if (!rdev_set_badblocks( | |
2486 | rdev, | |
2487 | r10_bio->devs[m].addr, | |
2488 | r10_bio->sectors, 0)) | |
2489 | md_error(conf->mddev, rdev); | |
749c55e9 | 2490 | } |
9ad1aefc N |
2491 | rdev = conf->mirrors[dev].replacement; |
2492 | if (r10_bio->devs[m].repl_bio == NULL) | |
2493 | continue; | |
2494 | if (test_bit(BIO_UPTODATE, | |
2495 | &r10_bio->devs[m].repl_bio->bi_flags)) { | |
2496 | rdev_clear_badblocks( | |
2497 | rdev, | |
2498 | r10_bio->devs[m].addr, | |
2499 | r10_bio->sectors); | |
2500 | } else { | |
2501 | if (!rdev_set_badblocks( | |
2502 | rdev, | |
2503 | r10_bio->devs[m].addr, | |
2504 | r10_bio->sectors, 0)) | |
2505 | md_error(conf->mddev, rdev); | |
2506 | } | |
1a0b7cd8 | 2507 | } |
749c55e9 N |
2508 | put_buf(r10_bio); |
2509 | } else { | |
bd870a16 N |
2510 | for (m = 0; m < conf->copies; m++) { |
2511 | int dev = r10_bio->devs[m].devnum; | |
2512 | struct bio *bio = r10_bio->devs[m].bio; | |
2513 | rdev = conf->mirrors[dev].rdev; | |
2514 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2515 | rdev_clear_badblocks( |
2516 | rdev, | |
2517 | r10_bio->devs[m].addr, | |
2518 | r10_bio->sectors); | |
2519 | rdev_dec_pending(rdev, conf->mddev); | |
bd870a16 N |
2520 | } else if (bio != NULL && |
2521 | !test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
2522 | if (!narrow_write_error(r10_bio, m)) { | |
2523 | md_error(conf->mddev, rdev); | |
2524 | set_bit(R10BIO_Degraded, | |
2525 | &r10_bio->state); | |
2526 | } | |
2527 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2528 | } |
475b0321 N |
2529 | bio = r10_bio->devs[m].repl_bio; |
2530 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2531 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2532 | rdev_clear_badblocks( |
2533 | rdev, | |
2534 | r10_bio->devs[m].addr, | |
2535 | r10_bio->sectors); | |
2536 | rdev_dec_pending(rdev, conf->mddev); | |
2537 | } | |
bd870a16 N |
2538 | } |
2539 | if (test_bit(R10BIO_WriteError, | |
2540 | &r10_bio->state)) | |
2541 | close_write(r10_bio); | |
749c55e9 N |
2542 | raid_end_bio_io(r10_bio); |
2543 | } | |
2544 | } | |
2545 | ||
fd01b88c | 2546 | static void raid10d(struct mddev *mddev) |
1da177e4 | 2547 | { |
9f2c9d12 | 2548 | struct r10bio *r10_bio; |
1da177e4 | 2549 | unsigned long flags; |
e879a879 | 2550 | struct r10conf *conf = mddev->private; |
1da177e4 | 2551 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2552 | struct blk_plug plug; |
1da177e4 LT |
2553 | |
2554 | md_check_recovery(mddev); | |
1da177e4 | 2555 | |
e1dfa0a2 | 2556 | blk_start_plug(&plug); |
1da177e4 | 2557 | for (;;) { |
6cce3b23 | 2558 | |
7eaceacc | 2559 | flush_pending_writes(conf); |
6cce3b23 | 2560 | |
a35e63ef N |
2561 | spin_lock_irqsave(&conf->device_lock, flags); |
2562 | if (list_empty(head)) { | |
2563 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2564 | break; |
a35e63ef | 2565 | } |
9f2c9d12 | 2566 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2567 | list_del(head->prev); |
4443ae10 | 2568 | conf->nr_queued--; |
1da177e4 LT |
2569 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2570 | ||
2571 | mddev = r10_bio->mddev; | |
070ec55d | 2572 | conf = mddev->private; |
bd870a16 N |
2573 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2574 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2575 | handle_write_completed(conf, r10_bio); |
2576 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) | |
1da177e4 | 2577 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2578 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2579 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2580 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2581 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2582 | else { |
2583 | /* just a partial read to be scheduled from a | |
2584 | * separate context | |
2585 | */ | |
2586 | int slot = r10_bio->read_slot; | |
2587 | generic_make_request(r10_bio->devs[slot].bio); | |
2588 | } | |
560f8e55 | 2589 | |
1d9d5241 | 2590 | cond_resched(); |
de393cde N |
2591 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2592 | md_check_recovery(mddev); | |
1da177e4 | 2593 | } |
e1dfa0a2 | 2594 | blk_finish_plug(&plug); |
1da177e4 LT |
2595 | } |
2596 | ||
2597 | ||
e879a879 | 2598 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2599 | { |
2600 | int buffs; | |
69335ef3 | 2601 | int i; |
1da177e4 LT |
2602 | |
2603 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2604 | BUG_ON(conf->r10buf_pool); |
69335ef3 N |
2605 | conf->have_replacement = 0; |
2606 | for (i = 0; i < conf->raid_disks; i++) | |
2607 | if (conf->mirrors[i].replacement) | |
2608 | conf->have_replacement = 1; | |
1da177e4 LT |
2609 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2610 | if (!conf->r10buf_pool) | |
2611 | return -ENOMEM; | |
2612 | conf->next_resync = 0; | |
2613 | return 0; | |
2614 | } | |
2615 | ||
2616 | /* | |
2617 | * perform a "sync" on one "block" | |
2618 | * | |
2619 | * We need to make sure that no normal I/O request - particularly write | |
2620 | * requests - conflict with active sync requests. | |
2621 | * | |
2622 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2623 | * that can be installed to exclude normal IO requests. | |
2624 | * | |
2625 | * Resync and recovery are handled very differently. | |
2626 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2627 | * | |
2628 | * For resync, we iterate over virtual addresses, read all copies, | |
2629 | * and update if there are differences. If only one copy is live, | |
2630 | * skip it. | |
2631 | * For recovery, we iterate over physical addresses, read a good | |
2632 | * value for each non-in_sync drive, and over-write. | |
2633 | * | |
2634 | * So, for recovery we may have several outstanding complex requests for a | |
2635 | * given address, one for each out-of-sync device. We model this by allocating | |
2636 | * a number of r10_bio structures, one for each out-of-sync device. | |
2637 | * As we setup these structures, we collect all bio's together into a list | |
2638 | * which we then process collectively to add pages, and then process again | |
2639 | * to pass to generic_make_request. | |
2640 | * | |
2641 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2642 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2643 | * has its remaining count decremented to 0, the whole complex operation | |
2644 | * is complete. | |
2645 | * | |
2646 | */ | |
2647 | ||
fd01b88c | 2648 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, |
ab9d47e9 | 2649 | int *skipped, int go_faster) |
1da177e4 | 2650 | { |
e879a879 | 2651 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2652 | struct r10bio *r10_bio; |
1da177e4 LT |
2653 | struct bio *biolist = NULL, *bio; |
2654 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2655 | int i; |
6cce3b23 | 2656 | int max_sync; |
57dab0bd | 2657 | sector_t sync_blocks; |
1da177e4 LT |
2658 | sector_t sectors_skipped = 0; |
2659 | int chunks_skipped = 0; | |
2660 | ||
2661 | if (!conf->r10buf_pool) | |
2662 | if (init_resync(conf)) | |
57afd89f | 2663 | return 0; |
1da177e4 LT |
2664 | |
2665 | skipped: | |
58c0fed4 | 2666 | max_sector = mddev->dev_sectors; |
1da177e4 LT |
2667 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
2668 | max_sector = mddev->resync_max_sectors; | |
2669 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2670 | /* If we aborted, we need to abort the |
2671 | * sync on the 'current' bitmap chucks (there can | |
2672 | * be several when recovering multiple devices). | |
2673 | * as we may have started syncing it but not finished. | |
2674 | * We can find the current address in | |
2675 | * mddev->curr_resync, but for recovery, | |
2676 | * we need to convert that to several | |
2677 | * virtual addresses. | |
2678 | */ | |
2679 | if (mddev->curr_resync < max_sector) { /* aborted */ | |
2680 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2681 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2682 | &sync_blocks, 1); | |
2683 | else for (i=0; i<conf->raid_disks; i++) { | |
2684 | sector_t sect = | |
2685 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2686 | bitmap_end_sync(mddev->bitmap, sect, | |
2687 | &sync_blocks, 1); | |
2688 | } | |
9ad1aefc N |
2689 | } else { |
2690 | /* completed sync */ | |
2691 | if ((!mddev->bitmap || conf->fullsync) | |
2692 | && conf->have_replacement | |
2693 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2694 | /* Completed a full sync so the replacements | |
2695 | * are now fully recovered. | |
2696 | */ | |
2697 | for (i = 0; i < conf->raid_disks; i++) | |
2698 | if (conf->mirrors[i].replacement) | |
2699 | conf->mirrors[i].replacement | |
2700 | ->recovery_offset | |
2701 | = MaxSector; | |
2702 | } | |
6cce3b23 | 2703 | conf->fullsync = 0; |
9ad1aefc | 2704 | } |
6cce3b23 | 2705 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2706 | close_sync(conf); |
57afd89f | 2707 | *skipped = 1; |
1da177e4 LT |
2708 | return sectors_skipped; |
2709 | } | |
2710 | if (chunks_skipped >= conf->raid_disks) { | |
2711 | /* if there has been nothing to do on any drive, | |
2712 | * then there is nothing to do at all.. | |
2713 | */ | |
57afd89f N |
2714 | *skipped = 1; |
2715 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2716 | } |
2717 | ||
c6207277 N |
2718 | if (max_sector > mddev->resync_max) |
2719 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2720 | ||
1da177e4 LT |
2721 | /* make sure whole request will fit in a chunk - if chunks |
2722 | * are meaningful | |
2723 | */ | |
2724 | if (conf->near_copies < conf->raid_disks && | |
2725 | max_sector > (sector_nr | conf->chunk_mask)) | |
2726 | max_sector = (sector_nr | conf->chunk_mask) + 1; | |
2727 | /* | |
2728 | * If there is non-resync activity waiting for us then | |
2729 | * put in a delay to throttle resync. | |
2730 | */ | |
0a27ec96 | 2731 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2732 | msleep_interruptible(1000); |
1da177e4 LT |
2733 | |
2734 | /* Again, very different code for resync and recovery. | |
2735 | * Both must result in an r10bio with a list of bios that | |
2736 | * have bi_end_io, bi_sector, bi_bdev set, | |
2737 | * and bi_private set to the r10bio. | |
2738 | * For recovery, we may actually create several r10bios | |
2739 | * with 2 bios in each, that correspond to the bios in the main one. | |
2740 | * In this case, the subordinate r10bios link back through a | |
2741 | * borrowed master_bio pointer, and the counter in the master | |
2742 | * includes a ref from each subordinate. | |
2743 | */ | |
2744 | /* First, we decide what to do and set ->bi_end_io | |
2745 | * To end_sync_read if we want to read, and | |
2746 | * end_sync_write if we will want to write. | |
2747 | */ | |
2748 | ||
6cce3b23 | 2749 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2750 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2751 | /* recovery... the complicated one */ | |
e875ecea | 2752 | int j; |
1da177e4 LT |
2753 | r10_bio = NULL; |
2754 | ||
ab9d47e9 N |
2755 | for (i=0 ; i<conf->raid_disks; i++) { |
2756 | int still_degraded; | |
9f2c9d12 | 2757 | struct r10bio *rb2; |
ab9d47e9 N |
2758 | sector_t sect; |
2759 | int must_sync; | |
e875ecea | 2760 | int any_working; |
24afd80d N |
2761 | struct mirror_info *mirror = &conf->mirrors[i]; |
2762 | ||
2763 | if ((mirror->rdev == NULL || | |
2764 | test_bit(In_sync, &mirror->rdev->flags)) | |
2765 | && | |
2766 | (mirror->replacement == NULL || | |
2767 | test_bit(Faulty, | |
2768 | &mirror->replacement->flags))) | |
ab9d47e9 | 2769 | continue; |
1da177e4 | 2770 | |
ab9d47e9 N |
2771 | still_degraded = 0; |
2772 | /* want to reconstruct this device */ | |
2773 | rb2 = r10_bio; | |
2774 | sect = raid10_find_virt(conf, sector_nr, i); | |
24afd80d N |
2775 | /* Unless we are doing a full sync, or a replacement |
2776 | * we only need to recover the block if it is set in | |
2777 | * the bitmap | |
ab9d47e9 N |
2778 | */ |
2779 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2780 | &sync_blocks, 1); | |
2781 | if (sync_blocks < max_sync) | |
2782 | max_sync = sync_blocks; | |
2783 | if (!must_sync && | |
24afd80d | 2784 | mirror->replacement == NULL && |
ab9d47e9 N |
2785 | !conf->fullsync) { |
2786 | /* yep, skip the sync_blocks here, but don't assume | |
2787 | * that there will never be anything to do here | |
2788 | */ | |
2789 | chunks_skipped = -1; | |
2790 | continue; | |
2791 | } | |
6cce3b23 | 2792 | |
ab9d47e9 N |
2793 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2794 | raise_barrier(conf, rb2 != NULL); | |
2795 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 2796 | |
ab9d47e9 N |
2797 | r10_bio->master_bio = (struct bio*)rb2; |
2798 | if (rb2) | |
2799 | atomic_inc(&rb2->remaining); | |
2800 | r10_bio->mddev = mddev; | |
2801 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
2802 | r10_bio->sector = sect; | |
1da177e4 | 2803 | |
ab9d47e9 N |
2804 | raid10_find_phys(conf, r10_bio); |
2805 | ||
2806 | /* Need to check if the array will still be | |
2807 | * degraded | |
2808 | */ | |
2809 | for (j=0; j<conf->raid_disks; j++) | |
2810 | if (conf->mirrors[j].rdev == NULL || | |
2811 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
2812 | still_degraded = 1; | |
87fc767b | 2813 | break; |
1da177e4 | 2814 | } |
ab9d47e9 N |
2815 | |
2816 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2817 | &sync_blocks, still_degraded); | |
2818 | ||
e875ecea | 2819 | any_working = 0; |
ab9d47e9 | 2820 | for (j=0; j<conf->copies;j++) { |
e875ecea | 2821 | int k; |
ab9d47e9 | 2822 | int d = r10_bio->devs[j].devnum; |
5e570289 | 2823 | sector_t from_addr, to_addr; |
3cb03002 | 2824 | struct md_rdev *rdev; |
40c356ce N |
2825 | sector_t sector, first_bad; |
2826 | int bad_sectors; | |
ab9d47e9 N |
2827 | if (!conf->mirrors[d].rdev || |
2828 | !test_bit(In_sync, &conf->mirrors[d].rdev->flags)) | |
2829 | continue; | |
2830 | /* This is where we read from */ | |
e875ecea | 2831 | any_working = 1; |
40c356ce N |
2832 | rdev = conf->mirrors[d].rdev; |
2833 | sector = r10_bio->devs[j].addr; | |
2834 | ||
2835 | if (is_badblock(rdev, sector, max_sync, | |
2836 | &first_bad, &bad_sectors)) { | |
2837 | if (first_bad > sector) | |
2838 | max_sync = first_bad - sector; | |
2839 | else { | |
2840 | bad_sectors -= (sector | |
2841 | - first_bad); | |
2842 | if (max_sync > bad_sectors) | |
2843 | max_sync = bad_sectors; | |
2844 | continue; | |
2845 | } | |
2846 | } | |
ab9d47e9 N |
2847 | bio = r10_bio->devs[0].bio; |
2848 | bio->bi_next = biolist; | |
2849 | biolist = bio; | |
2850 | bio->bi_private = r10_bio; | |
2851 | bio->bi_end_io = end_sync_read; | |
2852 | bio->bi_rw = READ; | |
5e570289 | 2853 | from_addr = r10_bio->devs[j].addr; |
24afd80d N |
2854 | bio->bi_sector = from_addr + rdev->data_offset; |
2855 | bio->bi_bdev = rdev->bdev; | |
2856 | atomic_inc(&rdev->nr_pending); | |
2857 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
2858 | |
2859 | for (k=0; k<conf->copies; k++) | |
2860 | if (r10_bio->devs[k].devnum == i) | |
2861 | break; | |
2862 | BUG_ON(k == conf->copies); | |
5e570289 | 2863 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 2864 | r10_bio->devs[0].devnum = d; |
5e570289 | 2865 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 2866 | r10_bio->devs[1].devnum = i; |
5e570289 | 2867 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 2868 | |
24afd80d N |
2869 | rdev = mirror->rdev; |
2870 | if (!test_bit(In_sync, &rdev->flags)) { | |
2871 | bio = r10_bio->devs[1].bio; | |
2872 | bio->bi_next = biolist; | |
2873 | biolist = bio; | |
2874 | bio->bi_private = r10_bio; | |
2875 | bio->bi_end_io = end_sync_write; | |
2876 | bio->bi_rw = WRITE; | |
2877 | bio->bi_sector = to_addr | |
2878 | + rdev->data_offset; | |
2879 | bio->bi_bdev = rdev->bdev; | |
2880 | atomic_inc(&r10_bio->remaining); | |
2881 | } else | |
2882 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
2883 | ||
2884 | /* and maybe write to replacement */ | |
2885 | bio = r10_bio->devs[1].repl_bio; | |
2886 | if (bio) | |
2887 | bio->bi_end_io = NULL; | |
2888 | rdev = mirror->replacement; | |
2889 | /* Note: if rdev != NULL, then bio | |
2890 | * cannot be NULL as r10buf_pool_alloc will | |
2891 | * have allocated it. | |
2892 | * So the second test here is pointless. | |
2893 | * But it keeps semantic-checkers happy, and | |
2894 | * this comment keeps human reviewers | |
2895 | * happy. | |
2896 | */ | |
2897 | if (rdev == NULL || bio == NULL || | |
2898 | test_bit(Faulty, &rdev->flags)) | |
2899 | break; | |
2900 | bio->bi_next = biolist; | |
2901 | biolist = bio; | |
2902 | bio->bi_private = r10_bio; | |
2903 | bio->bi_end_io = end_sync_write; | |
2904 | bio->bi_rw = WRITE; | |
2905 | bio->bi_sector = to_addr + rdev->data_offset; | |
2906 | bio->bi_bdev = rdev->bdev; | |
2907 | atomic_inc(&r10_bio->remaining); | |
ab9d47e9 N |
2908 | break; |
2909 | } | |
2910 | if (j == conf->copies) { | |
e875ecea N |
2911 | /* Cannot recover, so abort the recovery or |
2912 | * record a bad block */ | |
ab9d47e9 N |
2913 | put_buf(r10_bio); |
2914 | if (rb2) | |
2915 | atomic_dec(&rb2->remaining); | |
2916 | r10_bio = rb2; | |
e875ecea N |
2917 | if (any_working) { |
2918 | /* problem is that there are bad blocks | |
2919 | * on other device(s) | |
2920 | */ | |
2921 | int k; | |
2922 | for (k = 0; k < conf->copies; k++) | |
2923 | if (r10_bio->devs[k].devnum == i) | |
2924 | break; | |
24afd80d N |
2925 | if (!test_bit(In_sync, |
2926 | &mirror->rdev->flags) | |
2927 | && !rdev_set_badblocks( | |
2928 | mirror->rdev, | |
2929 | r10_bio->devs[k].addr, | |
2930 | max_sync, 0)) | |
2931 | any_working = 0; | |
2932 | if (mirror->replacement && | |
2933 | !rdev_set_badblocks( | |
2934 | mirror->replacement, | |
e875ecea N |
2935 | r10_bio->devs[k].addr, |
2936 | max_sync, 0)) | |
2937 | any_working = 0; | |
2938 | } | |
2939 | if (!any_working) { | |
2940 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
2941 | &mddev->recovery)) | |
2942 | printk(KERN_INFO "md/raid10:%s: insufficient " | |
2943 | "working devices for recovery.\n", | |
2944 | mdname(mddev)); | |
24afd80d | 2945 | mirror->recovery_disabled |
e875ecea N |
2946 | = mddev->recovery_disabled; |
2947 | } | |
ab9d47e9 | 2948 | break; |
1da177e4 | 2949 | } |
ab9d47e9 | 2950 | } |
1da177e4 LT |
2951 | if (biolist == NULL) { |
2952 | while (r10_bio) { | |
9f2c9d12 N |
2953 | struct r10bio *rb2 = r10_bio; |
2954 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
2955 | rb2->master_bio = NULL; |
2956 | put_buf(rb2); | |
2957 | } | |
2958 | goto giveup; | |
2959 | } | |
2960 | } else { | |
2961 | /* resync. Schedule a read for every block at this virt offset */ | |
2962 | int count = 0; | |
6cce3b23 | 2963 | |
78200d45 N |
2964 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
2965 | ||
6cce3b23 N |
2966 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
2967 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
2968 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
2969 | &mddev->recovery)) { | |
6cce3b23 N |
2970 | /* We can skip this block */ |
2971 | *skipped = 1; | |
2972 | return sync_blocks + sectors_skipped; | |
2973 | } | |
2974 | if (sync_blocks < max_sync) | |
2975 | max_sync = sync_blocks; | |
1da177e4 LT |
2976 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2977 | ||
1da177e4 LT |
2978 | r10_bio->mddev = mddev; |
2979 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
2980 | raise_barrier(conf, 0); |
2981 | conf->next_resync = sector_nr; | |
1da177e4 LT |
2982 | |
2983 | r10_bio->master_bio = NULL; | |
2984 | r10_bio->sector = sector_nr; | |
2985 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
2986 | raid10_find_phys(conf, r10_bio); | |
2987 | r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1; | |
2988 | ||
2989 | for (i=0; i<conf->copies; i++) { | |
2990 | int d = r10_bio->devs[i].devnum; | |
40c356ce N |
2991 | sector_t first_bad, sector; |
2992 | int bad_sectors; | |
2993 | ||
9ad1aefc N |
2994 | if (r10_bio->devs[i].repl_bio) |
2995 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
2996 | ||
1da177e4 LT |
2997 | bio = r10_bio->devs[i].bio; |
2998 | bio->bi_end_io = NULL; | |
af03b8e4 | 2999 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1da177e4 | 3000 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 3001 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 | 3002 | continue; |
40c356ce N |
3003 | sector = r10_bio->devs[i].addr; |
3004 | if (is_badblock(conf->mirrors[d].rdev, | |
3005 | sector, max_sync, | |
3006 | &first_bad, &bad_sectors)) { | |
3007 | if (first_bad > sector) | |
3008 | max_sync = first_bad - sector; | |
3009 | else { | |
3010 | bad_sectors -= (sector - first_bad); | |
3011 | if (max_sync > bad_sectors) | |
3012 | max_sync = max_sync; | |
3013 | continue; | |
3014 | } | |
3015 | } | |
1da177e4 LT |
3016 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
3017 | atomic_inc(&r10_bio->remaining); | |
3018 | bio->bi_next = biolist; | |
3019 | biolist = bio; | |
3020 | bio->bi_private = r10_bio; | |
3021 | bio->bi_end_io = end_sync_read; | |
802ba064 | 3022 | bio->bi_rw = READ; |
40c356ce | 3023 | bio->bi_sector = sector + |
1da177e4 LT |
3024 | conf->mirrors[d].rdev->data_offset; |
3025 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
3026 | count++; | |
9ad1aefc N |
3027 | |
3028 | if (conf->mirrors[d].replacement == NULL || | |
3029 | test_bit(Faulty, | |
3030 | &conf->mirrors[d].replacement->flags)) | |
3031 | continue; | |
3032 | ||
3033 | /* Need to set up for writing to the replacement */ | |
3034 | bio = r10_bio->devs[i].repl_bio; | |
3035 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
3036 | ||
3037 | sector = r10_bio->devs[i].addr; | |
3038 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
3039 | bio->bi_next = biolist; | |
3040 | biolist = bio; | |
3041 | bio->bi_private = r10_bio; | |
3042 | bio->bi_end_io = end_sync_write; | |
3043 | bio->bi_rw = WRITE; | |
3044 | bio->bi_sector = sector + | |
3045 | conf->mirrors[d].replacement->data_offset; | |
3046 | bio->bi_bdev = conf->mirrors[d].replacement->bdev; | |
3047 | count++; | |
1da177e4 LT |
3048 | } |
3049 | ||
3050 | if (count < 2) { | |
3051 | for (i=0; i<conf->copies; i++) { | |
3052 | int d = r10_bio->devs[i].devnum; | |
3053 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3054 | rdev_dec_pending(conf->mirrors[d].rdev, |
3055 | mddev); | |
9ad1aefc N |
3056 | if (r10_bio->devs[i].repl_bio && |
3057 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3058 | rdev_dec_pending( | |
3059 | conf->mirrors[d].replacement, | |
3060 | mddev); | |
1da177e4 LT |
3061 | } |
3062 | put_buf(r10_bio); | |
3063 | biolist = NULL; | |
3064 | goto giveup; | |
3065 | } | |
3066 | } | |
3067 | ||
3068 | for (bio = biolist; bio ; bio=bio->bi_next) { | |
3069 | ||
3070 | bio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
3071 | if (bio->bi_end_io) | |
3072 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
3073 | bio->bi_vcnt = 0; | |
3074 | bio->bi_idx = 0; | |
3075 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
3076 | bio->bi_size = 0; |
3077 | } | |
3078 | ||
3079 | nr_sectors = 0; | |
6cce3b23 N |
3080 | if (sector_nr + max_sync < max_sector) |
3081 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3082 | do { |
3083 | struct page *page; | |
3084 | int len = PAGE_SIZE; | |
1da177e4 LT |
3085 | if (sector_nr + (len>>9) > max_sector) |
3086 | len = (max_sector - sector_nr) << 9; | |
3087 | if (len == 0) | |
3088 | break; | |
3089 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 3090 | struct bio *bio2; |
1da177e4 | 3091 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
3092 | if (bio_add_page(bio, page, len, 0)) |
3093 | continue; | |
3094 | ||
3095 | /* stop here */ | |
3096 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
3097 | for (bio2 = biolist; | |
3098 | bio2 && bio2 != bio; | |
3099 | bio2 = bio2->bi_next) { | |
3100 | /* remove last page from this bio */ | |
3101 | bio2->bi_vcnt--; | |
3102 | bio2->bi_size -= len; | |
3103 | bio2->bi_flags &= ~(1<< BIO_SEG_VALID); | |
1da177e4 | 3104 | } |
ab9d47e9 | 3105 | goto bio_full; |
1da177e4 LT |
3106 | } |
3107 | nr_sectors += len>>9; | |
3108 | sector_nr += len>>9; | |
3109 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
3110 | bio_full: | |
3111 | r10_bio->sectors = nr_sectors; | |
3112 | ||
3113 | while (biolist) { | |
3114 | bio = biolist; | |
3115 | biolist = biolist->bi_next; | |
3116 | ||
3117 | bio->bi_next = NULL; | |
3118 | r10_bio = bio->bi_private; | |
3119 | r10_bio->sectors = nr_sectors; | |
3120 | ||
3121 | if (bio->bi_end_io == end_sync_read) { | |
3122 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
3123 | generic_make_request(bio); | |
3124 | } | |
3125 | } | |
3126 | ||
57afd89f N |
3127 | if (sectors_skipped) |
3128 | /* pretend they weren't skipped, it makes | |
3129 | * no important difference in this case | |
3130 | */ | |
3131 | md_done_sync(mddev, sectors_skipped, 1); | |
3132 | ||
1da177e4 LT |
3133 | return sectors_skipped + nr_sectors; |
3134 | giveup: | |
3135 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3136 | * drives must be failed or in resync, all drives |
3137 | * have a bad block, so try the next chunk... | |
1da177e4 | 3138 | */ |
09b4068a N |
3139 | if (sector_nr + max_sync < max_sector) |
3140 | max_sector = sector_nr + max_sync; | |
3141 | ||
3142 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3143 | chunks_skipped ++; |
3144 | sector_nr = max_sector; | |
1da177e4 | 3145 | goto skipped; |
1da177e4 LT |
3146 | } |
3147 | ||
80c3a6ce | 3148 | static sector_t |
fd01b88c | 3149 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3150 | { |
3151 | sector_t size; | |
e879a879 | 3152 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3153 | |
3154 | if (!raid_disks) | |
84707f38 | 3155 | raid_disks = conf->raid_disks; |
80c3a6ce | 3156 | if (!sectors) |
dab8b292 | 3157 | sectors = conf->dev_sectors; |
80c3a6ce DW |
3158 | |
3159 | size = sectors >> conf->chunk_shift; | |
3160 | sector_div(size, conf->far_copies); | |
3161 | size = size * raid_disks; | |
3162 | sector_div(size, conf->near_copies); | |
3163 | ||
3164 | return size << conf->chunk_shift; | |
3165 | } | |
3166 | ||
dab8b292 | 3167 | |
e879a879 | 3168 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3169 | { |
e879a879 | 3170 | struct r10conf *conf = NULL; |
c93983bf | 3171 | int nc, fc, fo; |
1da177e4 | 3172 | sector_t stride, size; |
dab8b292 | 3173 | int err = -EINVAL; |
1da177e4 | 3174 | |
f73ea873 MT |
3175 | if (mddev->new_chunk_sectors < (PAGE_SIZE >> 9) || |
3176 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
128595ed N |
3177 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
3178 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
3179 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3180 | goto out; |
1da177e4 | 3181 | } |
2604b703 | 3182 | |
f73ea873 MT |
3183 | nc = mddev->new_layout & 255; |
3184 | fc = (mddev->new_layout >> 8) & 255; | |
3185 | fo = mddev->new_layout & (1<<16); | |
dab8b292 | 3186 | |
1da177e4 | 3187 | if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || |
f73ea873 | 3188 | (mddev->new_layout >> 17)) { |
128595ed | 3189 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 3190 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
3191 | goto out; |
3192 | } | |
dab8b292 TM |
3193 | |
3194 | err = -ENOMEM; | |
e879a879 | 3195 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3196 | if (!conf) |
1da177e4 | 3197 | goto out; |
dab8b292 | 3198 | |
4443ae10 | 3199 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
dab8b292 TM |
3200 | GFP_KERNEL); |
3201 | if (!conf->mirrors) | |
3202 | goto out; | |
4443ae10 N |
3203 | |
3204 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3205 | if (!conf->tmppage) | |
dab8b292 TM |
3206 | goto out; |
3207 | ||
1da177e4 | 3208 | |
64a742bc | 3209 | conf->raid_disks = mddev->raid_disks; |
1da177e4 LT |
3210 | conf->near_copies = nc; |
3211 | conf->far_copies = fc; | |
3212 | conf->copies = nc*fc; | |
c93983bf | 3213 | conf->far_offset = fo; |
dab8b292 TM |
3214 | conf->chunk_mask = mddev->new_chunk_sectors - 1; |
3215 | conf->chunk_shift = ffz(~mddev->new_chunk_sectors); | |
3216 | ||
3217 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, | |
3218 | r10bio_pool_free, conf); | |
3219 | if (!conf->r10bio_pool) | |
3220 | goto out; | |
3221 | ||
58c0fed4 | 3222 | size = mddev->dev_sectors >> conf->chunk_shift; |
64a742bc N |
3223 | sector_div(size, fc); |
3224 | size = size * conf->raid_disks; | |
3225 | sector_div(size, nc); | |
3226 | /* 'size' is now the number of chunks in the array */ | |
3227 | /* calculate "used chunks per device" in 'stride' */ | |
3228 | stride = size * conf->copies; | |
af03b8e4 N |
3229 | |
3230 | /* We need to round up when dividing by raid_disks to | |
3231 | * get the stride size. | |
3232 | */ | |
3233 | stride += conf->raid_disks - 1; | |
64a742bc | 3234 | sector_div(stride, conf->raid_disks); |
dab8b292 TM |
3235 | |
3236 | conf->dev_sectors = stride << conf->chunk_shift; | |
64a742bc | 3237 | |
c93983bf | 3238 | if (fo) |
64a742bc N |
3239 | stride = 1; |
3240 | else | |
c93983bf | 3241 | sector_div(stride, fc); |
64a742bc N |
3242 | conf->stride = stride << conf->chunk_shift; |
3243 | ||
1da177e4 | 3244 | |
e7e72bf6 | 3245 | spin_lock_init(&conf->device_lock); |
dab8b292 TM |
3246 | INIT_LIST_HEAD(&conf->retry_list); |
3247 | ||
3248 | spin_lock_init(&conf->resync_lock); | |
3249 | init_waitqueue_head(&conf->wait_barrier); | |
3250 | ||
3251 | conf->thread = md_register_thread(raid10d, mddev, NULL); | |
3252 | if (!conf->thread) | |
3253 | goto out; | |
3254 | ||
dab8b292 TM |
3255 | conf->mddev = mddev; |
3256 | return conf; | |
3257 | ||
3258 | out: | |
128595ed | 3259 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", |
dab8b292 TM |
3260 | mdname(mddev)); |
3261 | if (conf) { | |
3262 | if (conf->r10bio_pool) | |
3263 | mempool_destroy(conf->r10bio_pool); | |
3264 | kfree(conf->mirrors); | |
3265 | safe_put_page(conf->tmppage); | |
3266 | kfree(conf); | |
3267 | } | |
3268 | return ERR_PTR(err); | |
3269 | } | |
3270 | ||
fd01b88c | 3271 | static int run(struct mddev *mddev) |
dab8b292 | 3272 | { |
e879a879 | 3273 | struct r10conf *conf; |
dab8b292 | 3274 | int i, disk_idx, chunk_size; |
0f6d02d5 | 3275 | struct mirror_info *disk; |
3cb03002 | 3276 | struct md_rdev *rdev; |
dab8b292 TM |
3277 | sector_t size; |
3278 | ||
3279 | /* | |
3280 | * copy the already verified devices into our private RAID10 | |
3281 | * bookkeeping area. [whatever we allocate in run(), | |
3282 | * should be freed in stop()] | |
3283 | */ | |
3284 | ||
3285 | if (mddev->private == NULL) { | |
3286 | conf = setup_conf(mddev); | |
3287 | if (IS_ERR(conf)) | |
3288 | return PTR_ERR(conf); | |
3289 | mddev->private = conf; | |
3290 | } | |
3291 | conf = mddev->private; | |
3292 | if (!conf) | |
3293 | goto out; | |
3294 | ||
dab8b292 TM |
3295 | mddev->thread = conf->thread; |
3296 | conf->thread = NULL; | |
3297 | ||
8f6c2e4b MP |
3298 | chunk_size = mddev->chunk_sectors << 9; |
3299 | blk_queue_io_min(mddev->queue, chunk_size); | |
3300 | if (conf->raid_disks % conf->near_copies) | |
3301 | blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks); | |
3302 | else | |
3303 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3304 | (conf->raid_disks / conf->near_copies)); | |
3305 | ||
dafb20fa | 3306 | rdev_for_each(rdev, mddev) { |
34b343cf | 3307 | |
1da177e4 | 3308 | disk_idx = rdev->raid_disk; |
84707f38 | 3309 | if (disk_idx >= conf->raid_disks |
1da177e4 LT |
3310 | || disk_idx < 0) |
3311 | continue; | |
3312 | disk = conf->mirrors + disk_idx; | |
3313 | ||
56a2559b N |
3314 | if (test_bit(Replacement, &rdev->flags)) { |
3315 | if (disk->replacement) | |
3316 | goto out_free_conf; | |
3317 | disk->replacement = rdev; | |
3318 | } else { | |
3319 | if (disk->rdev) | |
3320 | goto out_free_conf; | |
3321 | disk->rdev = rdev; | |
3322 | } | |
3323 | ||
8f6c2e4b MP |
3324 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3325 | rdev->data_offset << 9); | |
1da177e4 LT |
3326 | |
3327 | disk->head_position = 0; | |
1da177e4 | 3328 | } |
6d508242 | 3329 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3330 | if (!enough(conf, -1)) { |
128595ed | 3331 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3332 | mdname(mddev)); |
1da177e4 LT |
3333 | goto out_free_conf; |
3334 | } | |
3335 | ||
3336 | mddev->degraded = 0; | |
3337 | for (i = 0; i < conf->raid_disks; i++) { | |
3338 | ||
3339 | disk = conf->mirrors + i; | |
3340 | ||
56a2559b N |
3341 | if (!disk->rdev && disk->replacement) { |
3342 | /* The replacement is all we have - use it */ | |
3343 | disk->rdev = disk->replacement; | |
3344 | disk->replacement = NULL; | |
3345 | clear_bit(Replacement, &disk->rdev->flags); | |
3346 | } | |
3347 | ||
5fd6c1dc | 3348 | if (!disk->rdev || |
2e333e89 | 3349 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3350 | disk->head_position = 0; |
3351 | mddev->degraded++; | |
8c2e870a NB |
3352 | if (disk->rdev) |
3353 | conf->fullsync = 1; | |
1da177e4 | 3354 | } |
d890fa2b | 3355 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3356 | } |
3357 | ||
8c6ac868 | 3358 | if (mddev->recovery_cp != MaxSector) |
128595ed | 3359 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
3360 | " -- starting background reconstruction\n", |
3361 | mdname(mddev)); | |
1da177e4 | 3362 | printk(KERN_INFO |
128595ed | 3363 | "md/raid10:%s: active with %d out of %d devices\n", |
84707f38 N |
3364 | mdname(mddev), conf->raid_disks - mddev->degraded, |
3365 | conf->raid_disks); | |
1da177e4 LT |
3366 | /* |
3367 | * Ok, everything is just fine now | |
3368 | */ | |
dab8b292 TM |
3369 | mddev->dev_sectors = conf->dev_sectors; |
3370 | size = raid10_size(mddev, 0, 0); | |
3371 | md_set_array_sectors(mddev, size); | |
3372 | mddev->resync_max_sectors = size; | |
1da177e4 | 3373 | |
0d129228 N |
3374 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
3375 | mddev->queue->backing_dev_info.congested_data = mddev; | |
7a5febe9 | 3376 | |
1da177e4 LT |
3377 | /* Calculate max read-ahead size. |
3378 | * We need to readahead at least twice a whole stripe.... | |
3379 | * maybe... | |
3380 | */ | |
3381 | { | |
9d8f0363 AN |
3382 | int stripe = conf->raid_disks * |
3383 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
1da177e4 LT |
3384 | stripe /= conf->near_copies; |
3385 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) | |
3386 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
3387 | } | |
3388 | ||
050b6615 | 3389 | blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); |
a91a2785 MP |
3390 | |
3391 | if (md_integrity_register(mddev)) | |
3392 | goto out_free_conf; | |
3393 | ||
1da177e4 LT |
3394 | return 0; |
3395 | ||
3396 | out_free_conf: | |
01f96c0a | 3397 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3398 | if (conf->r10bio_pool) |
3399 | mempool_destroy(conf->r10bio_pool); | |
1345b1d8 | 3400 | safe_put_page(conf->tmppage); |
990a8baf | 3401 | kfree(conf->mirrors); |
1da177e4 LT |
3402 | kfree(conf); |
3403 | mddev->private = NULL; | |
3404 | out: | |
3405 | return -EIO; | |
3406 | } | |
3407 | ||
fd01b88c | 3408 | static int stop(struct mddev *mddev) |
1da177e4 | 3409 | { |
e879a879 | 3410 | struct r10conf *conf = mddev->private; |
1da177e4 | 3411 | |
409c57f3 N |
3412 | raise_barrier(conf, 0); |
3413 | lower_barrier(conf); | |
3414 | ||
01f96c0a | 3415 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3416 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
3417 | if (conf->r10bio_pool) | |
3418 | mempool_destroy(conf->r10bio_pool); | |
990a8baf | 3419 | kfree(conf->mirrors); |
1da177e4 LT |
3420 | kfree(conf); |
3421 | mddev->private = NULL; | |
3422 | return 0; | |
3423 | } | |
3424 | ||
fd01b88c | 3425 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3426 | { |
e879a879 | 3427 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3428 | |
3429 | switch(state) { | |
3430 | case 1: | |
3431 | raise_barrier(conf, 0); | |
3432 | break; | |
3433 | case 0: | |
3434 | lower_barrier(conf); | |
3435 | break; | |
3436 | } | |
6cce3b23 | 3437 | } |
1da177e4 | 3438 | |
006a09a0 N |
3439 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3440 | { | |
3441 | /* Resize of 'far' arrays is not supported. | |
3442 | * For 'near' and 'offset' arrays we can set the | |
3443 | * number of sectors used to be an appropriate multiple | |
3444 | * of the chunk size. | |
3445 | * For 'offset', this is far_copies*chunksize. | |
3446 | * For 'near' the multiplier is the LCM of | |
3447 | * near_copies and raid_disks. | |
3448 | * So if far_copies > 1 && !far_offset, fail. | |
3449 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3450 | * multiply by chunk_size. Then round to this number. | |
3451 | * This is mostly done by raid10_size() | |
3452 | */ | |
3453 | struct r10conf *conf = mddev->private; | |
3454 | sector_t oldsize, size; | |
3455 | ||
3456 | if (conf->far_copies > 1 && !conf->far_offset) | |
3457 | return -EINVAL; | |
3458 | ||
3459 | oldsize = raid10_size(mddev, 0, 0); | |
3460 | size = raid10_size(mddev, sectors, 0); | |
3461 | md_set_array_sectors(mddev, size); | |
3462 | if (mddev->array_sectors > size) | |
3463 | return -EINVAL; | |
3464 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
3465 | revalidate_disk(mddev->gendisk); | |
3466 | if (sectors > mddev->dev_sectors && | |
3467 | mddev->recovery_cp > oldsize) { | |
3468 | mddev->recovery_cp = oldsize; | |
3469 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3470 | } | |
3471 | mddev->dev_sectors = sectors; | |
3472 | mddev->resync_max_sectors = size; | |
3473 | return 0; | |
3474 | } | |
3475 | ||
fd01b88c | 3476 | static void *raid10_takeover_raid0(struct mddev *mddev) |
dab8b292 | 3477 | { |
3cb03002 | 3478 | struct md_rdev *rdev; |
e879a879 | 3479 | struct r10conf *conf; |
dab8b292 TM |
3480 | |
3481 | if (mddev->degraded > 0) { | |
128595ed N |
3482 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
3483 | mdname(mddev)); | |
dab8b292 TM |
3484 | return ERR_PTR(-EINVAL); |
3485 | } | |
3486 | ||
dab8b292 TM |
3487 | /* Set new parameters */ |
3488 | mddev->new_level = 10; | |
3489 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3490 | mddev->new_layout = (1<<8) + 2; | |
3491 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3492 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3493 | mddev->raid_disks *= 2; |
3494 | /* make sure it will be not marked as dirty */ | |
3495 | mddev->recovery_cp = MaxSector; | |
3496 | ||
3497 | conf = setup_conf(mddev); | |
02214dc5 | 3498 | if (!IS_ERR(conf)) { |
dafb20fa | 3499 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3500 | if (rdev->raid_disk >= 0) |
3501 | rdev->new_raid_disk = rdev->raid_disk * 2; | |
02214dc5 KW |
3502 | conf->barrier = 1; |
3503 | } | |
3504 | ||
dab8b292 TM |
3505 | return conf; |
3506 | } | |
3507 | ||
fd01b88c | 3508 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3509 | { |
e373ab10 | 3510 | struct r0conf *raid0_conf; |
dab8b292 TM |
3511 | |
3512 | /* raid10 can take over: | |
3513 | * raid0 - providing it has only two drives | |
3514 | */ | |
3515 | if (mddev->level == 0) { | |
3516 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3517 | raid0_conf = mddev->private; |
3518 | if (raid0_conf->nr_strip_zones > 1) { | |
128595ed N |
3519 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
3520 | " with more than one zone.\n", | |
3521 | mdname(mddev)); | |
dab8b292 TM |
3522 | return ERR_PTR(-EINVAL); |
3523 | } | |
3524 | return raid10_takeover_raid0(mddev); | |
3525 | } | |
3526 | return ERR_PTR(-EINVAL); | |
3527 | } | |
3528 | ||
84fc4b56 | 3529 | static struct md_personality raid10_personality = |
1da177e4 LT |
3530 | { |
3531 | .name = "raid10", | |
2604b703 | 3532 | .level = 10, |
1da177e4 LT |
3533 | .owner = THIS_MODULE, |
3534 | .make_request = make_request, | |
3535 | .run = run, | |
3536 | .stop = stop, | |
3537 | .status = status, | |
3538 | .error_handler = error, | |
3539 | .hot_add_disk = raid10_add_disk, | |
3540 | .hot_remove_disk= raid10_remove_disk, | |
3541 | .spare_active = raid10_spare_active, | |
3542 | .sync_request = sync_request, | |
6cce3b23 | 3543 | .quiesce = raid10_quiesce, |
80c3a6ce | 3544 | .size = raid10_size, |
006a09a0 | 3545 | .resize = raid10_resize, |
dab8b292 | 3546 | .takeover = raid10_takeover, |
1da177e4 LT |
3547 | }; |
3548 | ||
3549 | static int __init raid_init(void) | |
3550 | { | |
2604b703 | 3551 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
3552 | } |
3553 | ||
3554 | static void raid_exit(void) | |
3555 | { | |
2604b703 | 3556 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
3557 | } |
3558 | ||
3559 | module_init(raid_init); | |
3560 | module_exit(raid_exit); | |
3561 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3562 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 3563 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 3564 | MODULE_ALIAS("md-raid10"); |
2604b703 | 3565 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
3566 | |
3567 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |